Radiomics-based brain aging prediction with multi-modal magnetic resonance imaging: detecting regional biomarkers in adults.

Article Figures and data Abstract Editor's evaluation eLife digest Introduction Methods Results Discussion Data availability References Decision letter Author response Article and author information Metrics Abstract Background: Estimates of ‘brain-predicted age’ quantify apparent brain age compared to normative trajectories of neuroimaging features. The brain age gap (BAG) between predicted and chronological age is elevated in symptomatic Alzheimer disease (AD) but has not been well explored in presymptomatic AD. Prior studies have typically modeled BAG with structural MRI, but more recently other modalities, including functional connectivity (FC) and multimodal MRI, have been explored. Methods: We trained three models to predict age from FC, structural (S), or multimodal MRI (S+FC) in 390 amyloid-negative cognitively normal (CN/A−) participants (18–89 years old). In independent samples of 144 CN/A−, 154 CN/A+, and 154 cognitively impaired (CI; CDR > 0) participants, we tested relationships between BAG and AD biomarkers of amyloid and tau, as well as a global cognitive composite. Results: All models predicted age in the control training set, with the multimodal model outperforming the unimodal models. All three BAG estimates were significantly elevated in CI compared to controls. FC-BAG was significantly reduced in CN/A+ participants compared to CN/A−. In CI participants only, elevated S-BAG and S+FC BAG were associated with more advanced AD pathology and lower cognitive performance. Conclusions: Both FC-BAG and S-BAG are elevated in CI participants. However, FC and structural MRI also capture complementary signals. Specifically, FC-BAG may capture a unique biphasic response to presymptomatic AD pathology, while S-BAG may capture pathological progression and cognitive decline in the symptomatic stage. A multimodal age-prediction model improves sensitivity to healthy age differences. Funding: This work was supported by the National Institutes of Health (P01-AG026276, P01- AG03991, P30-AG066444, 5-R01-AG052550, 5-R01-AG057680, 1-R01-AG067505, 1S10RR022984-01A1, and U19-AG032438), the BrightFocus Foundation (A2022014F), and the Alzheimer’s Association (SG-20-690363-DIAN). Editor's evaluation This is a useful study exploring multi-modality brain age (structural plus resting state MRI) in people in the early stages or at risk of Alzheimer's disease. They found solid evidence that people with cognitive impairment had older-appearing brains and that older-appearing brains were related to Alzheimer's risk factors such as amyloid and tau deposition. Their data suggest that the multi-modality brain age model is more accurate than a unimodal structural MRI model. https://doi.org/10.7554/eLife.81869.sa0 Decision letter Reviews on Sciety eLife's review process eLife digest The brains of people with advanced Alzheimer’s disease often look older than expected based on the patients’ actual age. This ‘brain age gap’ (how old a brain appears compared to the person’s chronological age) can be calculated thanks to machine learning algorithms which analyse images of the organ to detect changes related to aging. Traditionally, these models have relied on images of the brain structure, such as the size and thickness of various brain areas; more recent models have started to use activity data, such as how different brain regions work together to form functional networks. While the brain age gap is a useful measure for researchers who investigate aging and disease, it is not yet helpful for clinicians. For example, it is unclear whether the machine learning algorithm could detect changes in the brains of individuals in the initial stages of Alzheimer’s disease, before they start to manifest cognitive symptoms. Millar et al. explored this question by testing whether models which incorporate structural and activity data could be more sensitive to these early changes. Three machine learning algorithms (relying on either structural data, activity data, or combination of both) were used to predict the brain ages of participants with no sign of disease; with biological markers of Alzheimer’s disease but preserved cognitive functions; and with marked cognitive symptoms of the condition. Overall, the combined model was slightly better at predicting the brain age of healthy volunteers, and all three models indicated that patients with dementia had a brain which looked older than normal. For this group, the model based on structural data was also able to make predictions which reflected the severity of cognitive decline. Crucially, the algorithm which used activity data predicted that, in individuals with biological markers of Alzheimer’s disease but no cognitive impairment, the brain looked in fact younger than chronological age. Exactly why this is the case remains unclear, but this signal may be driven by neural processes which unfold in the early stages of the disease. While more research is needed, the work by Millar et al. helps to explore how various types of machine learning models could one day be used to assess and predict brain health. Introduction Alzheimer disease (AD) is marked by structural and functional disruptions in the brain, some of which can be observed through multimodal magnetic resonance imaging (MRI) in preclinical and symptomatic stages of the disease (Frisoni et al., 2010; Brier et al., 2014a). More recently, the ‘brain-predicted age’ framework has emerged as a promising tool for neuroimaging analyses, leveraging recent developments and accessibility of machine-learning techniques, as well as large-scale, publicly available neuroimaging datasets (Cole and Franke, 2017b; Franke and Gaser, 2019). These models are trained to quantify how ‘old’ a brain appears, as compared to a normative sample of training data - typically consisting of cognitively normal participants across the adult lifespan (e.g., Cole et al., 2015). Thus, the framework allows for a residual-based interpretation of the brain age gap (BAG), defined as the difference between model-predicted age and chronological age, as an index of vulnerability and/or resistance to underlying disease pathology. Indeed, several studies have demonstrated that BAG is elevated (i.e. the brain ‘appears older’ than expected) in a host of neurological and psychiatric disorders, including symptomatic AD (Franke et al., 2010; Franke and Gaser, 2012; Gaser et al., 2013), as well as schizophrenia (e.g., Koutsouleris et al., 2014), HIV (e.g., Cole et al., 2017c), and type-2 diabetes (e.g., Franke et al., 2013), and moreover, predicts mortality (Cole et al., 2018). Conversely, lower BAG is associated with lower risk of disease progression (Gaser et al., 2013; Wang et al., 2019; Bocancea et al., 2021). Critically, at least one comparison suggests that BAG exceeds other established MRI (hippocampal volume) and CSF (pTau and Aβ42) biomarkers in sensitivity to AD progression (Gaser et al., 2013). Thus, by summarizing complex, non-linear, highly multivariate patterns of neuroimaging features into a simple, interpretable summary metric, BAG may reflect a comprehensive biomarker of brain health. Several studies have established that symptomatic AD and mild cognitive impairment (MCI) are associated with elevated BAG (Cole and Franke, 2017b; Franke and Gaser, 2019). However, the sensitivity of these model estimates to AD in the presymptmatic stage (i.e. present amyloid pathology in the absence of cognitive decline [Sperling et al., 2011]) is less clear. The development of sensitive, reliable, non-invasive biomarkers of preclinical AD pathology is critical for the assessment of individual AD risk, as well as the evaluation of AD clinical prevention trials. Recent studies have demonstrated that greater BAG is associated with greater amyloid PET burden in a Down syndrome cohort (Cole et al., 2017a) and with greater tau PET burden in sporadic MCI and symptomatic AD (Lee et al., 2022). One approach to maximize sensitivity of BAG to presymptomatic AD pathology may be to train brain age models exclusively on amyloid-negative participants. As undetected AD pathology might influence MRI measures, and thus confound effects otherwise attributed to ‘healthy aging’ (Brier et al., 2014b), including the patterns learned by a traditional brain age model, an alternative model trained on amyloid-negative participants only might be more sensitive to detect presymptomatic AD pathology as deviations in BAG. Indeed, one recent study demonstrated that an amyloid-negative trained brain age model (Ly et al., 2020) is more sensitive to progressive stages of AD than a typical amyloid-insensitive model (Cole et al., 2015). However, this comparison included amyloid-negative and amyloid-positive test samples from two separate cohorts and thus may be driven by cohort, scanner, and/or site differences. To validate the applicability of the brain-predicted age approach to presymptomatic AD, it is important to test a model’s sensitivity to amyloid status, as well as continuous relationships with AD biomarkers, within a single cohort. Another recent comparison demonstrated that both traditional and amyloid-negative trained brain age models were similarly related to molecular AD biomarkers, but that further attempts to ‘disentangle’ AD from brain age by including more advanced AD continuum participants in the training sample significantly reduced relationships between brain age and AD markers (Hwang et al., 2022). Thus, in this study, we will apply the amyloid-negative training approach to a multimodal MRI dataset in order to maximize sensitivity to AD pathology in the presymptomatic stage. Most of the brain-predicted age reports described above focused primarily on structural MRI. However, other studies have successfully modeled brain age using a variety of other modalities, including metabolic PET (Goyal et al., 2019; Lee et al., 2022), diffusion MRI (Cherubini et al., 2016; Petersen et al., 2022), and functional connectivity (FC) (Dosenbach et al., 2010; Liem et al., 2017; Eavani et al., 2018; Nielsen et al., 2019). Integration of multiple neuroimaging modalities may maximize sensitivity of BAG estimates to preclinical AD. Indeed, recent multimodal comparisons suggest that structural MRI and FC capture complementary age-related signals (Eavani et al., 2018; Dunås et al., 2021) and that age prediction may be improved by incorporating multiple modalities (Liem et al., 2017; Engemann et al., 2020). One recent study has shown that BAG estimates from an FC graph theory-based model are significantly elevated in autosomal dominant AD mutation carriers and are positively associated with amyloid PET (Gonneaud et al., 2021). Furthermore, we have recently demonstrated that FC correlation-based BAG estimates are surprisingly reduced in cognitively normal participants with evidence of amyloid pathology and elevated pTau, as well as in cognitively normal APOE ε4 carriers at genetic risk of AD (Millar et al., 2022). Thus, incorporating FC into BAG models may improve sensitivity to early AD. This project aimed to develop multimodal models of brain-predicted age, incorporating both FC and structural MRI. Participants with presymptomatic AD pathology were excluded from the training set to maximize sensitivity. We hypothesized that BAG estimates would be sensitive to the presence of AD biomarkers and early cognitive impairment. We further considered whether estimates were continuously associated with AD biomarkers of amyloid and tau, as well as cognition. We hypothesized that FC and structural MRI would capture complementary signals related to age and AD. Thus, we systematically compared models trained on unimodal FC, structural MRI, and combined modalities to test the added utility of multimodal integration in accurately predicting age and whether each modality captures unique relationships with AD biomarkers and cognition. Methods Participants We formed a training sample of healthy controls spanning the adult lifespan by combining structural and FC-MRI data from three sources, as described previously (Millar et al., 2022): the Charles F. and Joanne Knight AD Research Center (ADRC) at Washington University in St. Louis (WUSTL), healthy controls from studies in the Ances lab at WUSTL (Thomas et al., 2013; Petersen et al., 2021), and mutation-negative controls from the Dominantly Inherited Alzheimer Network (DIAN) study of autosomal dominant AD at multiple international sites including WUSTL (McKay et al., 2022). To minimize the likelihood of undetected AD pathology in our training set, participants over the age of 50 were only included in the training set if they were cognitively normal, as assessed by the Clinical Dementia Rating (CDR 0; Morris, 1993), and had at least one biomarker indicating the absence of amyloid pathology (CN/A−, see below). We excluded 59 participants who did not have available CDR or biomarker measures (see Figure 1—figure supplement 1). As CDR and amyloid biomarkers were not available in the Ances lab controls, we included only participants at or below age 50 from this cohort in the training set. These healthy control participants were randomly divided into a training set (~80%; N=390) and a held-out test set (~20%; N=97), which did not significantly differ in age, sex, education, or race, see Table 1. Table 1 Demographic information of the combined samples. MeasureTraining sets (total N=390)Test sets (total N=97) §Analysis sets (total N=452)Ances Controls(CN/<50)DIAN Controls(CN/A−)Knight ADRC Controls(CN/A−)Ances Controls(CN/<50)DIAN Controls(CN/A−)Knight ADRC Controls(CN/A−)CN/A−CN/A+CIN136120134382633144154154Age (mean, SD)29.92 (9.92)40.02 (10.26)64.97 (10.57)26.68 (7.11)41.46 (12.34)64.73 (10.57)66.93 (8.53)72.56 (7.15)‡75.67 (6.86) ‡CDR (N 0 / N 0.5 / N 1.0 / N 2.0)NA120 / 0 / 0 / 0134 / 0 / 0 / 0NA26 / 0 / 0 / 033 / 0 / 0 / 0144 / 0 / 0 / 0154 / 0 / 0 / 00 / 119 / 35 / 2Amyloid status (N + / N -)NA120 / 0134 / 0NA26 / 033 / 0144 / 00 / 1540 / 57Biomarkers available (N PET / CSF / both)NA30 / 6 / 7911 / 22 / 91NA3 / 1 / 215 / 0 / 2824 / 0 / 12017 / 0 / 13714 / 0 / 43APOE ε4 carrier status (N + / N -)NA76 / 4499 / 34NA19 / 728 / 5115 / 2971 / 83 ‡55 / 98 ‡MMSE (mean, SD)NANA29.26 (1.05)NANA29.45 (0.94)29.13 (1.17)28.97 (1.33)25.37 (3.55) ‡Sex (N female / N male)70 / 6485 / 3584 / 5019 / 1816 / 1022 / 1189 / 5591 / 6368 / 86†Years of education (mean, SD)13.68 (2.16)14.78 (3.04)16.16 (2.43)13.95 (1.99)14.92 (2.83)16.48 (2.43)15.71 (2.65)15.90 (2.64)15.05 (2.97)*Race (N American Indian or Alaska Native)100100000Race (N Asian)112000010Race (N Black)670201707171620Race (N Native Hawaiian or Other Pacifc Islander)200200000Race (N White)57118112172626127137134SiteWUSTLMultiple sitesWUSTLWUSTLMultiple sitesWUSTLWUSTLWUSTLWUSTLScannerSiemens TrioSiemens Trio / VerioSiemens Trio / BiographSiemens TrioSiemens Trio / VerioSiemens Trio / BiographSiemens Trio / BiographSiemens Trio / BiographSiemens Trio / BiographField strength3T3T3T3T3T3T3T3T3T CN = Cognitively Normal, <50 = less than age 50, A− = amyloid negative, A+ = amyloid positive, CI = cognitively Impaired, DIAN = Dominantly Inherited Alzheimer Network, ADRC = Alzheimer Disease Research Center, AD = Alzheimer disease, CDR = Clinical Dementia Rating, MMSE = Mini Mental State Examination, WUSTL = Washington University in St. Louis, T = Tesla. Group differences from the CN/A− analysis set were tested with t tests for continuous variables and χ2 tests for categorical variables. * p < 0.05, ^ p < 0.10. † p < 0.01. ‡ p < 0.001. § Test sets include randomly-selected, non-overlapping subsets of participants drawn from the same studies as the training sets. Finally, independent samples for hypothesis testing included three groups from the Knight ADRC: a randomly selected sample of 144 CN/A− controls who did not overlap with the training or testing sets, 154 CN/A+ participants, and 154 cognitively impaired (CI) participants (CDR > 0 with a biomarker measure consistent with amyloid pathology [see below] and/or a primary diagnosis of AD or uncertain dementia [McKhann et al., 2011]). See Table 1 for demographic details of each sample. All participants provided written informed consent in accordance with the Declaration of Helsinki and their local institutional review board. All procedures were approved by the Human Research Protection Office at WUSTL (IRB ID # 201204041). PET and CSF biomarkers Amyloid burden was imaged with PET using (11 C)-Pittsburgh Compound B (PIB; Klunk et al., 2004) or (18 F)-Florbetapir (AV45; Wong et al., 2010). Regional standard uptake ratios (SUVRs) were modeled from 30 to 60 min after injection for PIB and from 50 to 70 min for AV45, using cerebellar gray as the reference region (Su et al., 2013). Regions of interest were segmented automatically using FreeSurfer 5.3 (Fischl, 2012). Global amyloid burden was defined as the mean of partial-volume-corrected (PVC) SUVRs from bilateral precuneus, superior and rostral middle frontal, lateral and medial orbitofrontal, and superior and middle temporal regions (Su et al., 2013). Amyloid summary SUVRs were harmonized across tracers using a centiloid conversion (Su et al., 2018). Tau deposition was imaged with PET using (18 F)-Flortaucipir (AV-1451; Chien et al., 2013). Regional SUVRs were modeled from 80 to 100 min after injection, using cerebellar gray as the reference region. A tau summary measure was defined in the mean PVC SUVRs from bilateral amygdala, entorhinal, inferior temporal, and lateral occipital regions (Mishra et al., 2017). CSF was collected via lumbar puncture using methods described previously (Fagan et al., 2006). After overnight fasting, 20–30 mL samples of CSF were collected, centrifuged, then aliquoted (500 µL) in polypropylene tubes, and stored at –80°C. CSF amyloid β peptide 42 (Aβ42), Aβ40, and phosphorylated tau-181 (pTau) were measured with automated Lumipulse immunoassays (Fujirebio, Malvern, PA, USA) using a single lot of assays for each analyte. Aβ42 and pTau estimates were each normalized for individual differences in CSF production rates by forming a ratio with Aβ40 as the denominator (Hansson et al., 2019; Guo et al., 2020). As pTau/Aβ40 was highly skewed, we applied a log transformation to these estimates before statistical analysis. Amyloid positivity was defined using previously published cutoffs for PIB (SUVR > 1.42; Vlassenko et al., 2016) or AV45 (SUVR > 1.19; Su et al., 2019). Additionally, the CSF Aβ42/Aβ40 ratio has been shown to be highly concordant with amyloid PET (positivity cutoff < 0.0673; Schindler et al., 2018; Volluz et al., 2021). Thus, participants were defined as amyloid-positive (for CN/A+ and CI groups) if they had either a PIB, AV45, or CSF Aβ42/Aβ40 ratio measure in the positive range. Participants with discordant positivity between PET and CSF estimates were defined as amyloid-positive. Cognitive battery Knight ADRC participants completed a 2 hr battery of cognitive tests. We examined global cognition by forming a composite of tasks across cognitive domains, including processing speed (Trail Making A; Schindler et al., 2018), executive function (Trail Making B; Schindler et al., 2018), semantic fluency (Animal Naming; Armitage, 1946), and episodic memory (Free and Cued Selective Reminding Test free recall score; Goodglass and Kaplan, et al., This composite has recently been used to study individual differences in cognition in the preclinical AD biomarkers and structural MRI et al., 2018), as well as functional MRI measures (Millar et al., 2021). MRI All MRI data were using a scanner, was a variety of models within and across As described previously (Millar et al., 2022), participants in the Knight ADRC and Ances lab studies completed one of two structural MRI by with = or = or = or = of = or 1 or with = = 1 and an imaging with = = = of = for two 6 min of DIAN participants completed a = = = of = et al., 2022). for the DIAN participants across sites and with the difference min of see 1 for summary of structural and functional MRI et al., 2022). FC and features All MRI data were using as described previously et al., 2010; Millar et al., 2022), including and of to a of et al., 2012). to an on independent samples of either younger or CN older was using a of the functional with the and was included in a single that a of the in As described previously et al., Millar et al., 2022), processing was to for Data based on estimates > and/or of > above To further minimize the influence of on FC estimates et al., in all we only included with < and > after data a temporal < < and including from FreeSurfer (Fischl, brain and as well as the of these signals. Finally, data were at data were across within a set of regions of interest in and cerebellar et al., 2020). For each we calculated the of the between all We then used the of each as features for predicting age. site and/or differences between samples might confound neuroimaging we harmonized FC using an approach et al., et al., which has previously been applied to FC data et al., 2018). MRI processing and features All images and structural through a with FreeSurfer 5.3 et al., 2012). processing included of and gray to a and of the based on the et al., 2006). and of and were and by a of trained research to (Su et al., 2013). We then used the thickness estimates from regions et al., with estimates from regions et al., as features for predicting age. We harmonized structural features across sites and using the same approach et al., et al., which has also been applied to structural MRI data et al., 2018). process As described previously (Millar et al., 2022), machine-learning were using the in 2021). We trained two process et al., 2004) each with a function to predict chronological age using harmonized MRI features or in the training set. The was within each model by a of from to using across 100 training The of for each model was found (see Figure 1—figure supplement and was applied for all of that model. All other were set to function = and = in the training set was assessed using via the the of the mean and between chronological age and the age predictions across the We then of the models to predict age in data by the trained models to the held-out test set of healthy controls. Finally, we applied the same models to separate analysis sets of 154 154 CN/A+, and 144 CN/A− controls to test our effects and individual difference models were each with a single model. The multimodal model was by the predictions from each unimodal model as features for training a model (Liem et al., 2017; Engemann et al., Dunås et al., 2021). For each we calculated BAG estimates as the difference between chronological age and age predictions from the unimodal FC model structural model and multimodal model To for observed in models et al., 2018; et al., 2019; et al., we included chronological age as a in all statistical tests of BAG (Cole et al., et al., 2018). However, to estimates of prediction et al., 2021), only age prediction were used for model in the training and test sets. analysis All statistical were in 2020). Demographic differences in the AD samples were tested with t tests for continuous variables and χ2 tests for categorical using CN/A− controls as a reference in brain age model were tested using test of difference between one between age and each model prediction of age. To for age-related in BAG et al., 2018; as previously we for age as a all statistical tests. Group differences in each BAG were tested using an test with t tests on BAG using a for multiple of were tested by of of of were tested with models tested the effects of cognitive impairment (CDR > 0 CDR 0) and amyloid positivity on BAG estimates from each model, for age sex, and years of the influence of on measures et al., 2012; et al., 2012; et al., we also included mean as an of in the FC and S+FC models. We tested continuous relationships with AD biomarkers and cognitive estimates using including the same demographic and the of amyloid biomarkers was reduced in the CN/A− we excluded these participants from models testing continuous amyloid were as Results and Demographic of the training sets, test sets, and analysis sets are in Table 1. CN/A+ participants were older = p < and more to be APOE ε4 carriers = p < than amyloid-negative controls. Furthermore, CI participants were older = p < more = p = more to be APOE ε4 carriers = p < and had years of education = p < and lower MMSE = p < than amyloid-negative controls. of model All models accurately predicted chronological age in the training sets, as

Article Figures and data Abstract Editor's evaluation eLife digest Introduction Methods Results Discussion Data availability References Decision letter Author response Article and author information Metrics Abstract Background: Estimates of ‘brain-predicted age’ quantify apparent brain age compared to normative trajectories of neuroimaging features. The brain age gap (BAG) between predicted and chronological age is elevated in symptomatic Alzheimer disease (AD) but has not been well explored in presymptomatic AD. Prior studies have typically modeled BAG with structural MRI, but more recently other modalities, including functional connectivity (FC) and multimodal MRI, have been explored. Methods: We trained three models to predict age from FC, structural (S), or multimodal MRI (S+FC) in 390 amyloid-negative cognitively normal (CN/A−) participants (18–89 years old). In independent samples of 144 CN/A−, 154 CN/A+, and 154 cognitively impaired (CI; CDR > 0) participants, we tested relationships between BAG and AD biomarkers of amyloid and tau, as well as a global cognitive composite. Results: All models predicted age in the control training set, with the multimodal model outperforming the unimodal models. All three BAG estimates were significantly elevated in CI compared to controls. FC-BAG was significantly reduced in CN/A+ participants compared to CN/A−. In CI participants only, elevated S-BAG and S+FC BAG were associated with more advanced AD pathology and lower cognitive performance. Conclusions: Both FC-BAG and S-BAG are elevated in CI participants. However, FC and structural MRI also capture complementary signals. Specifically, FC-BAG may capture a unique biphasic response to presymptomatic AD pathology, while S-BAG may capture pathological progression and cognitive decline in the symptomatic stage. A multimodal age-prediction model improves sensitivity to healthy age differences. Funding: This work was supported by the National Institutes of Health (P01-AG026276, P01- AG03991, P30-AG066444, 5-R01-AG052550, 5-R01-AG057680, 1-R01-AG067505, 1S10RR022984-01A1, and U19-AG032438), the BrightFocus Foundation (A2022014F), and the Alzheimer’s Association (SG-20-690363-DIAN). Editor's evaluation This is a useful study exploring multi-modality brain age (structural plus resting state MRI) in people in the early stages or at risk of Alzheimer's disease. They found solid evidence that people with cognitive impairment had older-appearing brains and that older-appearing brains were related to Alzheimer's risk factors such as amyloid and tau deposition. Their data suggest that the multi-modality brain age model is more accurate than a unimodal structural MRI model. https://doi.org/10.7554/eLife.81869.sa0 Decision letter Reviews on Sciety eLife's review process eLife digest The brains of people with advanced Alzheimer’s disease often look older than expected based on the patients’ actual age. This ‘brain age gap’ (how old a brain appears compared to the person’s chronological age) can be calculated thanks to machine learning algorithms which analyse images of the organ to detect changes related to aging. Traditionally, these models have relied on images of the brain structure, such as the size and thickness of various brain areas; more recent models have started to use activity data, such as how different brain regions work together to form functional networks. While the brain age gap is a useful measure for researchers who investigate aging and disease, it is not yet helpful for clinicians. For example, it is unclear whether the machine learning algorithm could detect changes in the brains of individuals in the initial stages of Alzheimer’s disease, before they start to manifest cognitive symptoms. Millar et al. explored this question by testing whether models which incorporate structural and activity data could be more sensitive to these early changes. Three machine learning algorithms (relying on either structural data, activity data, or combination of both) were used to predict the brain ages of participants with no sign of disease; with biological markers of Alzheimer’s disease but preserved cognitive functions; and with marked cognitive symptoms of the condition. Overall, the combined model was slightly better at predicting the brain age of healthy volunteers, and all three models indicated that patients with dementia had a brain which looked older than normal. For this group, the model based on structural data was also able to make predictions which reflected the severity of cognitive decline. Crucially, the algorithm which used activity data predicted that, in individuals with biological markers of Alzheimer’s disease but no cognitive impairment, the brain looked in fact younger than chronological age. Exactly why this is the case remains unclear, but this signal may be driven by neural processes which unfold in the early stages of the disease. While more research is needed, the work by Millar et al. helps to explore how various types of machine learning models could one day be used to assess and predict brain health. Introduction Alzheimer disease (AD) is marked by structural and functional disruptions in the brain, some of which can be observed through multimodal magnetic resonance imaging (MRI) in preclinical and symptomatic stages of the disease (Frisoni et al., 2010; Brier et al., 2014a). More recently, the ‘brain-predicted age’ framework has emerged as a promising tool for neuroimaging analyses, leveraging recent developments and accessibility of machine-learning techniques, as well as large-scale, publicly available neuroimaging datasets (Cole and Franke, 2017b; Franke and Gaser, 2019). These models are trained to quantify how ‘old’ a brain appears, as compared to a normative sample of training data - typically consisting of cognitively normal participants across the adult lifespan (e.g., Cole et al., 2015). Thus, the framework allows for a residual-based interpretation of the brain age gap (BAG), defined as the difference between model-predicted age and chronological age, as an index of vulnerability and/or resistance to underlying disease pathology. Indeed, several studies have demonstrated that BAG is elevated (i.e. the brain ‘appears older’ than expected) in a host of neurological and psychiatric disorders, including symptomatic AD (Franke et al., 2010; Franke and Gaser, 2012; Gaser et al., 2013), as well as schizophrenia (e.g., Koutsouleris et al., 2014), HIV (e.g., Cole et al., 2017c), and type-2 diabetes (e.g., Franke et al., 2013), and moreover, predicts mortality (Cole et al., 2018). Conversely, lower BAG is associated with lower risk of disease progression (Gaser et al., 2013; Wang et al., 2019; Bocancea et al., 2021). Critically, at least one comparison suggests that BAG exceeds other established MRI (hippocampal volume) and CSF (pTau and Aβ42) biomarkers in sensitivity to AD progression (Gaser et al., 2013). Thus, by summarizing complex, non-linear, highly multivariate patterns of neuroimaging features into a simple, interpretable summary metric, BAG may reflect a comprehensive biomarker of brain health. Several studies have established that symptomatic AD and mild cognitive impairment (MCI) are associated with elevated BAG (Cole and Franke, 2017b; Franke and Gaser, 2019). However, the sensitivity of these model estimates to AD in the presymptmatic stage (i.e. present amyloid pathology in the absence of cognitive decline [Sperling et al., 2011]) is less clear. The development of sensitive, reliable, non-invasive biomarkers of preclinical AD pathology is critical for the assessment of individual AD risk, as well as the evaluation of AD clinical prevention trials. Recent studies have demonstrated that greater BAG is associated with greater amyloid PET burden in a Down syndrome cohort (Cole et al., 2017a) and with greater tau PET burden in sporadic MCI and symptomatic AD (Lee et al., 2022). One approach to maximize sensitivity of BAG to presymptomatic AD pathology may be to train brain age models exclusively on amyloid-negative participants. As undetected AD pathology might influence MRI measures, and thus confound effects otherwise attributed to ‘healthy aging’ (Brier et al., 2014b), including the patterns learned by a traditional brain age model, an alternative model trained on amyloid-negative participants only might be more sensitive to detect presymptomatic AD pathology as deviations in BAG. Indeed, one recent study demonstrated that an amyloid-negative trained brain age model (Ly et al., 2020) is more sensitive to progressive stages of AD than a typical amyloid-insensitive model (Cole et al., 2015). However, this comparison included amyloid-negative and amyloid-positive test samples from two separate cohorts and thus may be driven by cohort, scanner, and/or site differences. To validate the applicability of the brain-predicted age approach to presymptomatic AD, it is important to test a model’s sensitivity to amyloid status, as well as continuous relationships with AD biomarkers, within a single cohort. Another recent comparison demonstrated that both traditional and amyloid-negative trained brain age models were similarly related to molecular AD biomarkers, but that further attempts to ‘disentangle’ AD from brain age by including more advanced AD continuum participants in the training sample significantly reduced relationships between brain age and AD markers (Hwang et al., 2022). Thus, in this study, we will apply the amyloid-negative training approach to a multimodal MRI dataset in order to maximize sensitivity to AD pathology in the presymptomatic stage. Most of the brain-predicted age reports described above focused primarily on structural MRI. However, other studies have successfully modeled brain age using a variety of other modalities, including metabolic PET (Goyal et al., 2019; Lee et al., 2022), diffusion MRI (Cherubini et al., 2016; Petersen et al., 2022), and functional connectivity (FC) (Dosenbach et al., 2010; Liem et al., 2017; Eavani et al., 2018; Nielsen et al., 2019). Integration of multiple neuroimaging modalities may maximize sensitivity of BAG estimates to preclinical AD. Indeed, recent multimodal comparisons suggest that structural MRI and FC capture complementary age-related signals (Eavani et al., 2018; Dunås et al., 2021) and that age prediction may be improved by incorporating multiple modalities (Liem et al., 2017; Engemann et al., 2020). One recent study has shown that BAG estimates from an FC graph theory-based model are significantly elevated in autosomal dominant AD mutation carriers and are positively associated with amyloid PET (Gonneaud et al., 2021). Furthermore, we have recently demonstrated that FC correlation-based BAG estimates are surprisingly reduced in cognitively normal participants with evidence of amyloid pathology and elevated pTau, as well as in cognitively normal APOE ε4 carriers at genetic risk of AD (Millar et al., 2022). Thus, incorporating FC into BAG models may improve sensitivity to early AD. This project aimed to develop multimodal models of brain-predicted age, incorporating both FC and structural MRI. Participants with presymptomatic AD pathology were excluded from the training set to maximize sensitivity. We hypothesized that BAG estimates would be sensitive to the presence of AD biomarkers and early cognitive impairment. We further considered whether estimates were continuously associated with AD biomarkers of amyloid and tau, as well as cognition. We hypothesized that FC and structural MRI would capture complementary signals related to age and AD. Thus, we systematically compared models trained on unimodal FC, structural MRI, and combined modalities to test the added utility of multimodal integration in accurately predicting age and whether each modality captures unique relationships with AD biomarkers and cognition. Methods Participants We formed a training sample of healthy controls spanning the adult lifespan by combining structural and FC-MRI data from three sources, as described previously (Millar et al., 2022): the Charles F. and Joanne Knight AD Research Center (ADRC) at Washington University in St. Louis (WUSTL), healthy controls from studies in the Ances lab at WUSTL (Thomas et al., 2013; Petersen et al., 2021), and mutation-negative controls from the Dominantly Inherited Alzheimer Network (DIAN) study of autosomal dominant AD at multiple international sites including WUSTL (McKay et al., 2022). To minimize the likelihood of undetected AD pathology in our training set, participants over the age of 50 were only included in the training set if they were cognitively normal, as assessed by the Clinical Dementia Rating (CDR 0; Morris, 1993), and had at least one biomarker indicating the absence of amyloid pathology (CN/A−, see below). We excluded 59 participants who did not have available CDR or biomarker measures (see Figure 1—figure supplement 1). As CDR and amyloid biomarkers were not available in the Ances lab controls, we included only participants at or below age 50 from this cohort in the training set. These healthy control participants were randomly divided into a training set (~80%; N=390) and a held-out test set (~20%; N=97), which did not significantly differ in age, sex, education, or race, see Table 1. Table 1 Demographic information of the combined samples. MeasureTraining sets (total N=390)Test sets (total N=97) §Analysis sets (total N=452)Ances Controls(CN/<50)DIAN Controls(CN/A−)Knight ADRC Controls(CN/A−)Ances Controls(CN/<50)DIAN Controls(CN/A−)Knight ADRC Controls(CN/A−)CN/A−CN/A+CIN136120134382633144154154Age (mean, SD)29.92 (9.92)40.02 (10.26)64.97 (10.57)26.68 (7.11)41.46 (12.34)64.73 (10.57)66.93 (8.53)72.56 (7.15)‡75.67 (6.86) ‡CDR (N 0 / N 0.5 / N 1.0 / N 2.0)NA120 / 0 / 0 / 0134 / 0 / 0 / 0NA26 / 0 / 0 / 033 / 0 / 0 / 0144 / 0 / 0 / 0154 / 0 / 0 / 00 / 119 / 35 / 2Amyloid status (N + / N -)NA120 / 0134 / 0NA26 / 033 / 0144 / 00 / 1540 / 57Biomarkers available (N PET / CSF / both)NA30 / 6 / 7911 / 22 / 91NA3 / 1 / 215 / 0 / 2824 / 0 / 12017 / 0 / 13714 / 0 / 43APOE ε4 carrier status (N + / N -)NA76 / 4499 / 34NA19 / 728 / 5115 / 2971 / 83 ‡55 / 98 ‡MMSE (mean, SD)NANA29.26 (1.05)NANA29.45 (0.94)29.13 (1.17)28.97 (1.33)25.37 (3.55) ‡Sex (N female / N male)70 / 6485 / 3584 / 5019 / 1816 / 1022 / 1189 / 5591 / 6368 / 86†Years of education (mean, SD)13.68 (2.16)14.78 (3.04)16.16 (2.43)13.95 (1.99)14.92 (2.83)16.48 (2.43)15.71 (2.65)15.90 (2.64)15.05 (2.97)*Race (N American Indian or Alaska Native)100100000Race (N Asian)112000010Race (N Black)670201707171620Race (N Native Hawaiian or Other Pacifc Islander)200200000Race (N White)57118112172626127137134SiteWUSTLMultiple sitesWUSTLWUSTLMultiple sitesWUSTLWUSTLWUSTLWUSTLScannerSiemens TrioSiemens Trio / VerioSiemens Trio / BiographSiemens TrioSiemens Trio / VerioSiemens Trio / BiographSiemens Trio / BiographSiemens Trio / BiographSiemens Trio / BiographField strength3T3T3T3T3T3T3T3T3T CN = Cognitively Normal, <50 = less than age 50, A− = amyloid negative, A+ = amyloid positive, CI = cognitively Impaired, DIAN = Dominantly Inherited Alzheimer Network, ADRC = Alzheimer Disease Research Center, AD = Alzheimer disease, CDR = Clinical Dementia Rating, MMSE = Mini Mental State Examination, WUSTL = Washington University in St. Louis, T = Tesla. Group differences from the CN/A− analysis set were tested with t tests for continuous variables and χ2 tests for categorical variables. * p < 0.05, ^ p < 0.10. † p < 0.01. ‡ p < 0.001. § Test sets include randomly-selected, non-overlapping subsets of participants drawn from the same studies as the training sets. Finally, independent samples for hypothesis testing included three groups from the Knight ADRC: a randomly selected sample of 144 CN/A− controls who did not overlap with the training or testing sets, 154 CN/A+ participants, and 154 cognitively impaired (CI) participants (CDR > 0 with a biomarker measure consistent with amyloid pathology [see below] and/or a primary diagnosis of AD or uncertain dementia [McKhann et al., 2011]). See Table 1 for demographic details of each sample. All participants provided written informed consent in accordance with the Declaration of Helsinki and their local institutional review board. All procedures were approved by the Human Research Protection Office at WUSTL (IRB ID # 201204041). PET and CSF biomarkers Amyloid burden was imaged with PET using (11 C)-Pittsburgh Compound B (PIB; Klunk et al., 2004) or (18 F)-Florbetapir (AV45; Wong et al., 2010). Regional standard uptake ratios (SUVRs) were modeled from 30 to 60 min after injection for PIB and from 50 to 70 min for AV45, using cerebellar gray as the reference region (Su et al., 2013). Regions of interest were segmented automatically using FreeSurfer 5.3 (Fischl, 2012). Global amyloid burden was defined as the mean of partial-volume-corrected (PVC) SUVRs from bilateral precuneus, superior and rostral middle frontal, lateral and medial orbitofrontal, and superior and middle temporal regions (Su et al., 2013). Amyloid summary SUVRs were harmonized across tracers using a centiloid conversion (Su et al., 2018). Tau deposition was imaged with PET using (18 F)-Flortaucipir (AV-1451; Chien et al., 2013). Regional SUVRs were modeled from 80 to 100 min after injection, using cerebellar gray as the reference region. A tau summary measure was defined in the mean PVC SUVRs from bilateral amygdala, entorhinal, inferior temporal, and lateral occipital regions (Mishra et al., 2017). CSF was collected via lumbar puncture using methods described previously (Fagan et al., 2006). After overnight fasting, 20–30 mL samples of CSF were collected, centrifuged, then aliquoted (500 µL) in polypropylene tubes, and stored at –80°C. CSF amyloid β peptide 42 (Aβ42), Aβ40, and phosphorylated tau-181 (pTau) were measured with automated Lumipulse immunoassays (Fujirebio, Malvern, PA, USA) using a single lot of assays for each analyte. Aβ42 and pTau estimates were each normalized for individual differences in CSF production rates by forming a ratio with Aβ40 as the denominator (Hansson et al., 2019; Guo et al., 2020). As pTau/Aβ40 was highly skewed, we applied a log transformation to these estimates before statistical analysis. Amyloid positivity was defined using previously published cutoffs for PIB (SUVR > 1.42; Vlassenko et al., 2016) or AV45 (SUVR > 1.19; Su et al., 2019). Additionally, the CSF Aβ42/Aβ40 ratio has been shown to be highly concordant with amyloid PET (positivity cutoff < 0.0673; Schindler et al., 2018; Volluz et al., 2021). Thus, participants were defined as amyloid-positive (for CN/A+ and CI groups) if they had either a PIB, AV45, or CSF Aβ42/Aβ40 ratio measure in the positive range. Participants with discordant positivity between PET and CSF estimates were defined as amyloid-positive. Cognitive battery Knight ADRC participants completed a 2 hr battery of cognitive tests. We examined global cognition by forming a composite of tasks across cognitive domains, including processing speed (Trail Making A; Schindler et al., 2018), executive function (Trail Making B; Schindler et al., 2018), semantic fluency (Animal Naming; Armitage, 1946), and episodic memory (Free and Cued Selective Reminding Test free recall score; Goodglass and Kaplan, et al., This composite has recently been used to study individual differences in cognition in the preclinical AD biomarkers and structural MRI et al., 2018), as well as functional MRI measures (Millar et al., 2021). MRI All MRI data were using a scanner, was a variety of models within and across As described previously (Millar et al., 2022), participants in the Knight ADRC and Ances lab studies completed one of two structural MRI by with = or = or = or = of = or 1 or with = = 1 and an imaging with = = = of = for two 6 min of DIAN participants completed a = = = of = et al., 2022). for the DIAN participants across sites and with the difference min of see 1 for summary of structural and functional MRI et al., 2022). FC and features All MRI data were using as described previously et al., 2010; Millar et al., 2022), including and of to a of et al., 2012). to an on independent samples of either younger or CN older was using a of the functional with the and was included in a single that a of the in As described previously et al., Millar et al., 2022), processing was to for Data based on estimates > and/or of > above To further minimize the influence of on FC estimates et al., in all we only included with < and > after data a temporal < < and including from FreeSurfer (Fischl, brain and as well as the of these signals. Finally, data were at data were across within a set of regions of interest in and cerebellar et al., 2020). For each we calculated the of the between all We then used the of each as features for predicting age. site and/or differences between samples might confound neuroimaging we harmonized FC using an approach et al., et al., which has previously been applied to FC data et al., 2018). MRI processing and features All images and structural through a with FreeSurfer 5.3 et al., 2012). processing included of and gray to a and of the based on the et al., 2006). and of and were and by a of trained research to (Su et al., 2013). We then used the thickness estimates from regions et al., with estimates from regions et al., as features for predicting age. We harmonized structural features across sites and using the same approach et al., et al., which has also been applied to structural MRI data et al., 2018). process As described previously (Millar et al., 2022), machine-learning were using the in 2021). We trained two process et al., 2004) each with a function to predict chronological age using harmonized MRI features or in the training set. The was within each model by a of from to using across 100 training The of for each model was found (see Figure 1—figure supplement and was applied for all of that model. All other were set to function = and = in the training set was assessed using via the the of the mean and between chronological age and the age predictions across the We then of the models to predict age in data by the trained models to the held-out test set of healthy controls. Finally, we applied the same models to separate analysis sets of 154 154 CN/A+, and 144 CN/A− controls to test our effects and individual difference models were each with a single model. The multimodal model was by the predictions from each unimodal model as features for training a model (Liem et al., 2017; Engemann et al., Dunås et al., 2021). For each we calculated BAG estimates as the difference between chronological age and age predictions from the unimodal FC model structural model and multimodal model To for observed in models et al., 2018; et al., 2019; et al., we included chronological age as a in all statistical tests of BAG (Cole et al., et al., 2018). However, to estimates of prediction et al., 2021), only age prediction were used for model in the training and test sets. analysis All statistical were in 2020). Demographic differences in the AD samples were tested with t tests for continuous variables and χ2 tests for categorical using CN/A− controls as a reference in brain age model were tested using test of difference between one between age and each model prediction of age. To for age-related in BAG et al., 2018; as previously we for age as a all statistical tests. Group differences in each BAG were tested using an test with t tests on BAG using a for multiple of were tested by of of of were tested with models tested the effects of cognitive impairment (CDR > 0 CDR 0) and amyloid positivity on BAG estimates from each model, for age sex, and years of the influence of on measures et al., 2012; et al., 2012; et al., we also included mean as an of in the FC and S+FC models. We tested continuous relationships with AD biomarkers and cognitive estimates using including the same demographic and the of amyloid biomarkers was reduced in the CN/A− we excluded these participants from models testing continuous amyloid were as Results and Demographic of the training sets, test sets, and analysis sets are in Table 1. CN/A+ participants were older = p < and more to be APOE ε4 carriers = p < than amyloid-negative controls. Furthermore, CI participants were older = p < more = p = more to be APOE ε4 carriers = p < and had years of education = p < and lower MMSE = p < than amyloid-negative controls. of model All models accurately predicted chronological age in the training sets, as

BackgroundResting heart rate (RHR), has been related to increased risk of dementia, but the relationship between RHR and brain age is unclear.ObjectiveWe aimed to investigate the association of RHR with brain age and brain age gap (BAG, the difference between predicted brain age and chronological age) assessed by multimodal Magnetic Resonance Imaging (MRI) in mid- and old-aged adults.DesignA longitudinal study from the UK Biobank neuroimaging project where participants underwent brain MRI scans 9+ years after baseline.SettingA population-based study.ParticipantsA total of 33,381 individuals (mean age 54.74 ± 7.49 years; 53.44% female).MeasurementsBaseline RHR was assessed by blood pressure monitor and categorized as <60, 60–69 (reference), 70–79, or ≥80 beats per minute (bpm). Brain age was predicted using LASSO through 1,079 phenotypes in six MRI modalities (including T1-weighted MRI, T2-FLAIR, T2*, diffusion-MRI, task fMRI, and resting-state fMRI). Data were analyzed using linear regression models.ResultsAs a continuous variable, higher RHR was associated with older brain age (β for per 1-SD increase: 0.331, 95% [95% confidence interval, CI]: 0.265, 0.398) and larger BAG (β: 0.263, 95% CI: 0.202, 0.324). As a categorical variable, RHR 70–79 bpm and RHR ≥80 bpm were associated with older brain age (β [95% CI]: 0.361 [0.196, 0.526] / 0.737 [0.517, 0.957]) and larger BAG (0.256 [0.105, 0.407] / 0.638 [0.436, 0.839]), but RHR< 60 bpm with younger brain age (−0.324 [−0.500, −0.147]) and smaller BAG (−0.230 [−0.392, −0.067]), compared to the reference group. These associations between elevated RHR and brain age were similar in both middle-aged (<60) and older (≥60) adults, whereas the association of RHR< 60 bpm with younger brain age and larger BAG was only significant among middle-aged adults. In stratification analysis, the association between RHR ≥80 bpm and older brain age was present in people with and without CVDs, while the relation of RHR 70–79 bpm to brain age present only in people with CVD.ConclusionHigher RHR (>80 bpm) is associated with older brain age, even among middle-aged adults, but RHR< 60 bpm is associated with younger brain age. Greater RHR could be an indicator for accelerated brain aging.

2037 Background: Predictors of overall survival (OS) after brain metastasis (BM) in HER2-positive breast cancer (BC) are not well characterized. This study aimed to identify clinical and imaging-derived (radiomic) features that predict OS and develop a combined model for better prognostic performance. Methods: Our retrospective study analyzed 289 patients initially diagnosed with non-metastatic HER2-positive BC who later developed BM. We used 25 clinical characteristics and 12 treatment parameters to develop a Clinical model. We developed an Imaging model using a subset of 120 patients, who possessed evaluable pre-treatment brain MRI for delineating tumor segmentations on all brain metastatic lesions. We extracted 1078 radiomic features from each tumor segmentation using PyRadiomics, generating 8 feature sets based on 2 segmentation strategies (largest tumor per patient versus all tumors combined) and 4 tumor feature types (entire tumor, solid component, necrotic component, combined solid and necrotic features with statistical transformations). Morphological features, including lesion number, total size/volume, and necrotic-to-solid ratios, were also incorporated, along with tumor intracranial location. Cox proportional hazards regression model with Coxnet, integrating LASSO and Elastic Net regularization, was used to predict OS. For fair comparison, we randomly selected 30% (n = 31) of the smallest subset (n = 103, largest brain metastasis with both necrotic and solid components), all of which overlap with other model subsets, as validation cohort. Three model types—Clinical, Imaging and Combined—were compared using the concordance index (C-index) to assess performance based on validation cohort. Results: Clinical model, built on the whole cohort (286 women, 3 men; mean age 54.52 ± 12.79 years), identified 3 predictors of OS. Imaging model, built on a subset of 120 patients with brain MRI data, identified a radiomic signature (RS) consisting of 4 radiomic features most predictive of OS. Using the same subset, the Combined model (C-index: 0.728 [95% CI: 0.590–0.855]) outperformed Clinical (C-index: 0.62 [95% CI: 0.44–0.78]) and Imaging (C-index: 0.62 [95% CI: 0.46–0.77]) models in the held-out validation cohort (n = 31). Significant features associated with increased mortality risk in the Combined model included a higher RS, absence of tucatinib treatment for the primary BC prior to BM development, elevated Ki-67 expression, Black race, higher N stage, and brainstem metastases. Among these factors, RS, with the largest absolute coefficient in the Combined model (0.38), emerged as the most important predictor of OS (hazard ratio: 20.03 [95% CI: 4.92–81.48], p &lt; 0.005). Conclusions: A distinct RS from brain MRI is the strongest predictor of OS in patients with BM from HER2-positive BC, surpassing clinical factors. RS may refine risk stratification and guide treatment or clinical trial prioritization.

Although many studies on brain-age prediction in patients with schizophrenia have been reported recently, none has predicted brain age based on different neuroimaging modalities and different brain regions in these patients. Here, we constructed brain-age prediction models with multimodal MRI and examined the deviations of aging trajectories in different brain regions of participants with schizophrenia recruited from multiple centers. The data of 230 healthy controls (HCs) were used for model training. Next, we investigated the differences in brain age gaps between participants with schizophrenia and HCs from two independent cohorts. A Gaussian process regression algorithm with fivefold cross-validation was used to train 90, 90, and 48 models for gray matter (GM), functional connectivity (FC), and fractional anisotropy (FA) maps in the training dataset, respectively. The brain age gaps in different brain regions for all participants were calculated, and the differences in brain age gaps between the two groups were examined. Our results showed that most GM regions in participants with schizophrenia in both cohorts exhibited accelerated aging, particularly in the frontal lobe, temporal lobe, and insula. The parts of the white matter tracts, including the cerebrum and cerebellum, indicated deviations in aging trajectories in participants with schizophrenia. However, no accelerated brain aging was noted in the FC maps. The accelerated aging in 22 GM regions and 10 white matter tracts in schizophrenia potentially exacerbates with disease progression. In individuals with schizophrenia, different brain regions demonstrate dynamic deviations of brain aging trajectories. Our findings provided more insights into schizophrenia neuropathology.

For a long history, the diagnosis of psychiatric disorders relies on behavioral and self-reported symptoms, lacking objective neural or imaging biomarkers. Recent advances in neuroimaging techniques open opportunities for objective diagnosis protocols in psychopathology. However, the formidable dimensionality of neuroimaging data and the complexity of their relation to psychiatric disorders pose great challenges to analytical methods. In this dissertation, we propose methods to improve both the prediction performance and consistency of biomarker detection of machine learning (ML) models utilizing multimodal neuroimaging data to assist in the diagnosis of psychiatric disorders. We also examine how altered brain development is related to psychiatric disorders and propose a multi-dimensional brain age prediction approach to achieve a more sensitive quantification of precocity and delay in brain development. We first investigate the reproducibility of feature selection of ML models with brain imaging data. Consistent selection of predictive neuroimaging features via ML plays an important role in improving our understanding of and ability to treat psychiatric disorders. We compute a reproducibility index (R-index) for each feature as the reciprocal of the coefficient of variation of absolute mean difference across a larger number of bootstrap samples. The R-index is then integrated into regularized classification models as penalty weight. Reproducible features with a larger R-index are assigned smaller penalty weights and thus are more likely to be selected by our proposed R-index regularized classification models. We expect the proposed models will result in better prediction accuracy and more consistent feature selection for both simulated and real brain imaging data. Results show that our proposed R-index models are effective in separating informative features from noise features. Additionally, the proposed models yield better prediction performance and coefficient estimation than the standard regularized classification models. Improvements achieved by the proposed models are essential to advance our understanding of the selected brain imaging features as well as their associations with psychiatric disorders. As psychiatric disorders frequently evolve during adolescence when drastic brain changes emerge, in the second project, we examine how brain development is related to psychiatric disorders and cognition. Brain age prediction using ML techniques has drawn great attention. The brain age gap, which is defined as the difference between the predicted age (brain age) and the chronological age, has been reported to associate with altered brain development in various psychiatric disorders and precocity or delay of cognition in a healthy population. We evaluate the performance of brain age prediction with 36 different combinations of imaging features and ML models including deep learning. Single and multimodal brain imaging data including MRI, DTI, and rs-fMRI from a large data set with 839 subjects are examined. Additionally, we explore the potential nonlinear relationship between the brain age gap and chronological age and propose a new approach to correct the systematic bias of the brain age gap. The result show that multi-modal brain imaging features improve prediction performance and that psychiatric disorders are vulnerable to altered brain development. Our study are also helpful to advance the practice of optimizing different analytic methodologies in brain age prediction. Lastly, we extend the brain age prediction approach outlined above to obtain a multi-dimensional brain age index. As different brain regions and sub-systems mature at different stages of the lifespan, a unidimensional brain age may not capture the diverged development trajectory of different brain subsystems. In contrast to unidimensional brain age prediction, our proposed multi-dimensional brain age index (MBAI) is helpful to quantify the brain development of multiple sub-systems that show staggered developmental pace and distinct patterns. In addition, brain imaging features clustered in one subgroup are more homogeneous. Prediction of brain age using a subset of homogeneous features helps to mitigate overfitting of the age prediction model and thus improves the estimates of the brain age gap. Furthermore, we investigate how the multi-dimensional brain age index is altered in psychiatric disorders using data from the Philadelphia Neurodevelopmental Cohort (PNC) study. Our results show that the MBAI provides a flexible analysis of region-specific brain-age changes that are invisible to unidimensional brain-age prediction methods. Importantly, brain ages computed from region-specific feature clusters contain complementary information and demonstrate differential ability to classify disorder groups (e.g., specific phobia, depression, ADHD) from healthy controls. Compared to unidimensional brain-age indices, we show that the MBAI is sensitive to alterations in brain structures and captures distinct regional change patterns which may serve as biomarkers that may contribute to our understanding of healthy and pathological brain development and to the characterization, diagnosis, and, potentially, treatment of various disorders.

We aimed to develop and validate a nomogram combining bi-regional radiomics features from multimodal magnetic resonance imaging (MRI) and clinicoradiological characteristics to preoperatively predict microvascular invasion (MVI) of hepatocellular carcinoma (HCC). A total of 267 HCC patients were divided into training (n=194) and validation (n=73) cohorts according to MRI data. Bi-regional features were extracted from whole tumors and peritumoral regions in multimodal MRI. The minimum redundancy maximum relevance (mRMR) algorithm was applied to select features and build signatures. The predictive performance of the optimal radiomics signature was further evaluated within subgroups defined by tumor size and alpha fetoprotein (AFP) level. Then, a radiomics nomogram including the optimal radiomics signature, radiographic descriptors, and clinical variables was developed using multivariable regression. The nomogram performance was evaluated based on its discrimination, calibration, and clinical utility. The fusion radiomics signature derived from triphasic dynamic contrast-enhanced (DCE) MR images can effectively classify MVI and non-MVI HCC patients, with an AUC of 0.784 (95% CI: 0.719-0.840) in the training cohort and 0.820 (95% CI: 0.713-0.900) in the validation cohort. The fusion radiomics signature also performed well in the subgroups defined by the two risk factors, respectively. The nomogram, consisting of the fusion radiomics signature, arterial peritumoral enhancement, and AFP level, outperformed the clinicoradiological prediction model in the validation cohort (AUCs: 0.858 vs. 0.729; P=0.022), fitting well in the calibration curves (P>0.05). Decision curves confirmed the clinical utility of the nomogram. The radiomics nomogram can serve as a visual predictive tool for MVI in HCCs, and thus assist clinicians in selecting optimal treatment strategies to improve clinical outcomes.

AAGP Session to Highlight Research on Resilience, Emotional Well-Being in Aging

To identify the role of radiomics texture features both within and outside the nodule in predicting (a) time to progression (TTP) and overall survival (OS) as well as (b) response to chemotherapy in patients with non-small cell lung cancer (NSCLC). Data in a total of 125 patients who had been treated with pemetrexed-based platinum doublet chemotherapy at Cleveland Clinic were retrospectively analyzed. The patients were divided randomly into two sets with the constraint that there were an equal number of responders and nonresponders in the training set. The training set comprised 53 patients with NSCLC, and the validation set comprised 72 patients. A machine learning classifier trained with radiomic texture features extracted from intra- and peritumoral regions of non-contrast-enhanced CT images was used to predict response to chemotherapy. The radiomic risk-score signature was generated by using least absolute shrinkage and selection operator with the Cox regression model; association of the radiomic signature with TTP and OS was also evaluated. A combination of radiomic features in conjunction with a quadratic discriminant analysis classifier yielded a mean maximum area under the receiver operating characteristic curve (AUC) of 0.82 ± 0.09 (standard deviation) in the training set and a corresponding AUC of 0.77 in the independent testing set. The radiomics signature was also significantly associated with TTP (hazard ratio [HR], 2.8; 95% confidence interval [CI]: 1.95, 4.00; P < .0001) and OS (HR, 2.35; 95% CI: 1.41, 3.94; P = .0011). Additionally, decision curve analysis demonstrated that in terms of clinical usefulness, the radiomics signature had a higher overall net benefit in prediction of high-risk patients to receive treatment than the clinicopathologic measurements. This study suggests that radiomic texture features extracted from within and around the nodule on baseline CT scans are (a) predictive of response to chemotherapy and (b) associated with TTP and OS for patients with NSCLC.© RSNA, 2019Supplemental material is available for this article.

ObjectiveTo investigate radiomics features extracted from PET and CT components of 18F-FDG PET/CT images integrating clinical factors and metabolic parameters of PET to predict progression-free survival (PFS) in advanced high-grade serous ovarian cancer (HGSOC).MethodsA total of 261 patients were finally enrolled in this study and randomly divided into training (n=182) and validation cohorts (n=79). The data of clinical features and metabolic parameters of PET were reviewed from hospital information system(HIS). All volumes of interest (VOIs) of PET/CT images were semi-automatically segmented with a threshold of 42% of maximal standard uptake value (SUVmax) in PET images. A total of 1700 (850×2) radiomics features were separately extracted from PET and CT components of PET/CT images. Then two radiomics signatures (RSs) were constructed by the least absolute shrinkage and selection operator (LASSO) method. The RSs of PET (PET_RS) and CT components(CT_RS) were separately divided into low and high RS groups according to the optimum cutoff value. The potential associations between RSs with PFS were assessed in training and validation cohorts based on the Log-rank test. Clinical features and metabolic parameters of PET images (PET_MP) with P-value <0.05 in univariate and multivariate Cox regression were combined with PET_RS and CT_RS to develop prediction nomograms (Clinical, Clinical+ PET_MP, Clinical+ PET_RS, Clinical+ CT_RS, Clinical+ PET_MP + PET_RS, Clinical+ PET_MP + CT_RS) by using multivariate Cox regression. The concordance index (C-index), calibration curve, and net reclassification improvement (NRI) was applied to evaluate the predictive performance of nomograms in training and validation cohorts.ResultsIn univariate Cox regression analysis, six clinical features were significantly associated with PFS. Ten PET radiomics features were selected by LASSO to construct PET_RS, and 1 CT radiomics features to construct CT_RS. PET_RS and CT_RS was significantly associated with PFS both in training (P <0.00 for both RSs) and validation cohorts (P=0.01 for both RSs). Because there was no PET_MP significantly associated with PFS in training cohorts. Only three models were constructed by 4 clinical features with P-value <0.05 in multivariate Cox regression and RSs (Clinical, Clinical+ PET_RS, Clinical+ CT_RS). Clinical+ PET_RS model showed higher prognostic performance than other models in training cohort (C-index=0.70, 95% CI 0.68-0.72) and validation cohort (C-index=0.70, 95% CI 0.66-0.74). Calibration curves of each model for prediction of 1-, 3-year PFS indicated Clinical +PET_RS model showed excellent agreements between estimated and the observed 1-, 3-outcomes. Compared to the basic clinical model, Clinical+ PET_MS model resulted in greater improvement in predictive performance in the validation cohort.ConclusionPET_RS can improve diagnostic accuracy and provide complementary prognostic information compared with the use of clinical factors alone or combined with CT_RS. The newly developed radiomics nomogram is an effective tool to predict PFS for patients with advanced HGSOC.

A review on brain age prediction models

Machine learning (ML) has transformed neuroimaging research by enabling accurate predictions and feature extraction from large datasets. In this study, we investigate the application of six ML algorithms (Lasso, relevance vector regression, support vector regression, extreme gradient boosting, category boost, and multilayer perceptron) to predict brain age for middle-aged and older adults, which is a crucial area of research in neuroimaging. Despite the plethora of proposed ML models, there is no clear consensus on how to achieve better performance in brain age prediction for this population. Our study stands out by evaluating the impact of both ML algorithms and image modalities on brain age prediction performance using a large cohort of cognitively normal adults aged 44.6 to 82.3 years old (N = 27,842) with six image modalities. We found that the predictive performance of brain age is more reliant on the image modalities used than the ML algorithms employed. Specifically, our study highlights the superior performance of T1-weighted MRI and diffusion-weighted imaging and demonstrates that multi-modality-based brain age prediction significantly enhances performance compared to unimodality. Moreover, we identified Lasso as the most accurate ML algorithm for predicting brain age, achieving the lowest mean absolute error in both single-modality and multi-modality predictions. Additionally, Lasso also ranked highest in a comprehensive evaluation of the relationship between BrainAGE and the five frequently mentioned BrainAGE-related factors. Notably, our study also shows that ensemble learning outperforms Lasso when computational efficiency is not a concern. Overall, our study provides valuable insights into the development of accurate and reliable brain age prediction models for middle-aged and older adults, with significant implications for clinical practice and neuroimaging research. Our findings highlight the importance of image modality selection and emphasize Lasso as a promising ML algorithm for brain age prediction.

Not only are the effects of cardiovascular risk factors such as high blood pressure and low fitness on executive functions and brain activations in older adults scarcely investigated, no fMRI study has investigated the combined effects of multiple risk factors on brain activations in older adults. This fMRI study examined the independent and combined effects of two cardiovascular risk factors, arterial plasticity, and physical fitness, on brain activations during task-switching in older adults. The effects of these two risk factors on age-related differences in activation between older and younger adults were also examined. Independently, low physical fitness and low arterial plasticity were related to reduced suppressions of occipital brain regions. The combined effects of these two risks on occipital regions were greater than the independent effects of either risk factor. Age-related overactivations in frontal cortex were observed in low fitness older adults. Brain-behavior correlation indicates that these frontal overactivations are maladaptive to older adults’ task performance. It is possible that the resulting effects of cardiovascular risks on the aging brain, especially the maladaptive overactivations of frontal brain regions by high risk older adults, contribute to often found posterior-anterior shift in aging (PASA) brain activations. Furthermore, observed age-related differences in brain activations during task-switching can be partially attributed to individual differences in cardiovascular risks among older adults.

PurposeThe present study aims to comprehensively investigate the prognostic value of a radiomic nomogram that integrates contrast-enhanced computed tomography (CECT) radiomic signature and clinicopathological parameters in kidney renal clear cell carcinoma (KIRC).MethodsA total of 136 and 78 KIRC patients from the training and validation cohorts were included in the retrospective study. The intraclass correlation coefficient (ICC) was used to assess reproducibility of radiomic feature extraction. Univariate Cox analysis and least absolute shrinkage and selection operator (LASSO) as well as multivariate Cox analysis were utilized to construct radiomic signature and clinical signature in the training cohort. A prognostic nomogram was established containing a radiomic signature and clinicopathological parameters by using a multivariate Cox analysis. The predictive ability of the nomogram [relative operating characteristic curve (ROC), concordance index (C-index), Hosmer–Lemeshow test, and calibration curve] was evaluated in the training cohort and validated in the validation cohort. Patients were split into high- and low-risk groups, and the Kaplan–Meier (KM) method was conducted to identify the forecasting ability of the established models. In addition, genes related with the radiomic risk score were determined by weighted correlation network analysis (WGCNA) and were used to conduct functional analysis.ResultsA total of 2,944 radiomic features were acquired from the tumor volumes of interest (VOIs) of CECT images. The radiomic signature, including ten selected features, and the clinical signature, including three selected clinical variables, showed good performance in the training and validation cohorts [area under the curve (AUC), 0.897 and 0.712 for the radiomic signature; 0.827 and 0.822 for the clinical signature, respectively]. The radiomic prognostic nomogram showed favorable performance and calibration in the training cohort (AUC, 0.896, C-index, 0.846), which was verified in the validation cohort (AUC, 0.768). KM curves indicated that the progression-free interval (PFI) time was dramatically shorter in the high-risk group than in the low-risk group. The functional analysis indicated that radiomic signature was significantly associated with T cell activation.ConclusionsThe nomogram combined with CECT radiomic and clinicopathological signatures exhibits excellent power in predicting the PFI of KIRC patients, which may aid in clinical management and prognostic evaluation of cancer patients.

Preoperative computed tomography enterography-based radiomics signature: A potential predictor of postoperative anastomotic recurrence in patients with Crohn’s disease