Multiparametric Magnetic Resonance Imaging Scans Research Articles

NIMG-46. MULTI-PARAMETRIC MRI RADIOMIC ANALYSIS USING DEEP LEARNING PREDICTS PERITUMORAL GLIOBLASTOMA INFILTRATION AND SUBSEQUENT RECURRENCE

Abstract PURPOSE Glioblastoma (GBM) is the most aggressive and infiltrative brain tumor with a very poor prognosis which hasn’t significantly improved in over 20 years. The almost 100% recurrence rate is due to cancer infiltration beyond the tumor margins currently being targeted by standard of care. Prior studies used traditional supervised machine learning have shown great promise in predicting tumor infiltration beyond these margins. We hypothesize that deep learning methods can further improve such predictive maps and guide intensive, yet targeted and personalized therapies. METHODS MRIs from a cohort of 109 de novo GBM patients were collected from Hospital of the University of Pennsylvania. All the patients incorporated in this study had pre-operative multi-parametric MRI scans including T1, T1Gd, T2, T2-FLAIR, ADC, and underwent surgical resection followed by standard chemoradiation therapy and had pathologically confirmed recurrence. A novel, automated deep learning method, informed by results of prior studies using support vector machines, was constructed to train a patch-based ensemble CNN to identify regions of peri-tumoral cancer infiltration. Leave-one-out was used to evaluate the predictive value of this method, against pathology-proven subsequent recurrence. RESULTS Probability maps, representing the likelihood of tumor infiltration and eventual recurrence, were binarized using threshold of 50% cutoff, compared with actual recurrence on post-recurrence MRI scans. The average cross-validated accuracy was 92%, specificity was 93%, patient-based sensitivity was 78%, and odds-ratio among all patients was 12.95 (Estimated “hot spots” were 12.95 times more likely to present recurrence in the future). CONCLUSIONS This study demonstrates that Multi-parametric MRI pattern analysis using CNN based network can successfully predict tumor infiltration in peritumoral region for Glioblastoma patients. These suggest that new intensive, yet precisely targeted treatments can be developed, guided by AI-driven predictive maps of infiltration.

NIMG-42. RADIOIMMUNOMIC SIGNATURES IN PEDIATRIC LOW-GRADE GLIOMA FOR NON-INVASIVE IDENTIFICATION OF IMMUNOTHERAPEUTIC TARGETS

Abstract INTRODUCTION Tumor immune microenvironment (TIME) plays a key role in response to anti-tumor therapies. Therefore, detailed characterization of TIME is crucial for patient stratification and enrollment onto emerging immunotherapies. As standard-of-care treatments are not feasible or confer devastating long-term morbidity for pediatric low-grade glioma (pLGG), non-invasive quantification of the TIME using radiomic analysis, may guide treatment decisions. On the largest available transcriptomic and imaging data in pLGGs, provided through Children’s Brain Tumor Network (CBTN), this data-driven study aims to discover immunological profiles of pLGG that are meaningfully related to biomarkers of tumor immune response, and radioimmunomic signatures that are predictive of immunological profiles. METHODS Enrichment of immune and stromal cell types and pathways in 502 subjects were inferred using xCell algorithm applied to transcriptomic data and the tumors were then clustered via consensus clustering approach. Tumor inflammation signature (TIS), as a biomarker of response to anti-PD-1 blockade, was compared across the immunological clusters. Quantitative radiomic features were extracted from the segmented tumor regions on multiparametric MRI scans of 155/502 patients. Cross-validated support vector machines with forward feature selection were trained on radiomic features using one-versus-the-rest approach to predict the immunological groups. RESULTS Three immunological clusters, i.e., immune-cold, -altered, and -hot were found, with different levels of enrichment in immune and microenvironment scores, CD4 T-cells, and macrophages. Significantly higher TIS values were found in immune-hot and -altered compared to immune-cold tumors (p= 1.2e-8 and p= 7e-16, respectively). Radioimmunomic models yielded average AUC of 0.75 for one-versus-the-rest differentiation of the immunological clusters. CONCLUSIONS The proposed analysis revealed three distinct immunological groups in pLGGs, with immune-hot tumors more likely to respond favorably to immunotherapies. Furthermore, non-invasive radioimmunomic signatures were shown to potentially stratify the patients based on their TIME, which upon further development and validation, could inform treatments for pLGGs.

BTDNet: A Multi-Modal Approach for Brain Tumor Radiogenomic Classification

Brain tumors pose significant health challenges worldwide, with glioblastoma being one of the most aggressive forms. The accurate determination of the O6-methylguanine-DNA methyltransferase (MGMT) promoter methylation status is crucial for personalized treatment strategies. However, traditional methods are labor-intensive and time-consuming. This paper proposes a novel multi-modal approach, BTDNet, that leverages multi-parametric MRI scans, including FLAIR, T1w, T1wCE, and T2 3D volumes, to predict the MGMT promoter methylation status. BTDNet’s main contribution involves addressing two main challenges: the variable volume lengths (i.e., each volume consists of a different number of slices) and the volume-level annotations (i.e., the whole 3D volume is annotated and not the independent slices that it consists of). BTDNet consists of four components: (i) data augmentation (which performs geometric transformations, convex combinations of data pairs, and test-time data augmentation); (ii) 3D analysis (which performs global analysis through a CNN-RNN); (iii) routing (which contains a mask layer that handles variable input feature lengths); and (iv) modality fusion (which effectively enhances data representation, reduces ambiguities, and mitigates data scarcity). The proposed method outperformed state-of-the-art methods in the RSNA-ASNR-MICCAI BraTS 2021 Challenge by at least 3.3% in terms of the F1 score, offering a promising avenue for enhancing brain tumor diagnosis and treatment.

Attention mechanism based multi-sequence MRI fusion improves prediction of response to neoadjuvant chemoradiotherapy in locally advanced rectal cancer

BackgroundAccurate prediction of response to neoadjuvant chemoradiotherapy (nCRT) is very important for treatment plan decision in locally advanced rectal cancer (LARC). The aim of this study was to investigate whether self-attention mechanism based multi-sequence fusion strategy applied to multiparametric magnetic resonance imaging (MRI) based deep learning or hand-crafted radiomics model construction can improve prediction of response to nCRT in LARC.MethodsThis retrospective analysis enrolled 422 consecutive patients with LARC who received nCRT before surgery at two hospitals. All patients underwent multiparametric MRI scans with three imaging sequences. Tumor regression grade (TRG) was used to assess the response of nCRT based on the resected specimen. Patients were separated into 2 groups: poor responders (TRG 2, 3) versus good responders (TRG 0, 1). A self-attention mechanism, namely channel attention, was applied to fuse the three sequence information for deep learning and radiomics models construction. For comparison, other two models without channel attention were also constructed. All models were developed in the same hospital and validated in the other hospital.ResultsThe deep learning model with channel attention mechanism achieved area under the curves (AUCs) of 0.898 in the internal validation cohort and 0.873 in the external validation cohort, which was the best performed model in all cohorts. More importantly, both the deep learning and radiomics model that applied channel attention mechanism performed better than those without channel attention mechanism.ConclusionsThe self-attention mechanism based multi-sequence fusion strategy can improve prediction of response to nCRT in LARC.

The Role of Multiparametric MRI and MRI–targeted Biopsy in the Diagnosis of Radiorecurrent Prostate Cancer: An Analysis from the FORECAST Trial

BackgroundThe role of multiparametric magnetic resonance imaging (MRI) for detecting recurrent prostate cancer after radiotherapy is unclear. ObjectiveTo evaluate MRI and MRI-targeted biopsies for detecting intraprostatic cancer recurrence and planning for salvage focal ablation. Design, setting, and participantsFOcal RECurrent Assessment and Salvage Treatment (FORECAST; NCT01883128) was a prospective cohort diagnostic study that recruited 181 patients with suspected radiorecurrence at six UK centres (2014 to 2018); 144 were included here. InterventionAll patients underwent MRI with 5 mm transperineal template mapping biopsies; 84 had additional MRI-targeted biopsies. MRI scans with Likert scores of 3 to 5 were deemed suspicious. Outcome measurements and statistical analysisFirst, the diagnostic accuracy of MRI was calculated. Second, the pathological characteristics of MRI-detected and MRI-undetected tumours were compared using the Wilcoxon rank sum test and chi-square test for trend. Third, four biopsy strategies involving an MRI-targeted biopsy alone and with systematic biopsies of one to two other quadrants were studied. Fisher’s exact test was used to compare MRI-targeted biopsy alone with the best other strategy for the number of patients with missed cancer and the number of patients with cancer harbouring additional tumours in unsampled quadrants. Analyses focused primarily on detecting cancer of any grade or length. Last, eligibility for focal therapy was evaluated for men with localised (≤T3bN0M0) radiorecurrent disease. Results and limitationsOf 144 patients, 111 (77%) had cancer detected on biopsy. MRI sensitivity and specificity at the patient level were 0.95 (95% confidence interval [CI] 0.92 to 0.99) and 0.21 (95% CI 0.07 to 0.35), respectively. At the prostate quadrant level, 258/576 (45%) quadrants had cancer detected on biopsy. Sensitivity and specificity were 0.66 (95% CI 0.59 to 0.73) and 0.54 (95% CI 0.46 to 0.62), respectively. At the quadrant level, compared with MRI-undetected tumours, MRI-detected tumours had longer maximum cancer core length (median difference 3 mm [7 vs 4 mm]; 95% CI 1 to 4 mm, p < 0.001) and a higher grade group (p = 0.002). Of the 84 men who also underwent an MRI-targeted biopsy, 73 (87%) had recurrent cancer diagnosed. Performing an MRI-targeted biopsy alone missed cancer in 5/73 patients (7%; 95% CI 3 to 15%); with additional systematic sampling of the other ipsilateral and contralateral posterior quadrants (strategy 4), 2/73 patients (3%; 95% CI 0 to 10%) would have had cancer missed (difference 4%; 95% CI –3 to 11%, p = 0.4). If an MRI-targeted biopsy alone was performed, 43/73 (59%; 95% CI 47 to 69%) patients with cancer would have harboured undetected additional tumours in unsampled quadrants. This reduced but only to 7/73 patients (10%; 95% CI 4 to 19%) with strategy 4 (difference 49%; 95% CI 36 to 62%, p < 0.0001). Of 73 patients, 43 (59%; 95% CI 47 to 69%) had localised radiorecurrent cancer suitable for a form of focal ablation. ConclusionsFor patients with recurrent prostate cancer after radiotherapy, MRI and MRI-targeted biopsy, with or without perilesional sampling, will diagnose cancer in the majority where present. MRI-undetected cancers, defined as Likert scores of 1 to 2, were found to be smaller and of lower grade. However, if salvage focal ablation is planned, an MRI-targeted biopsy alone is insufficient for prostate mapping; approximately three of five patients with recurrent cancer found on an MRI-targeted biopsy alone harboured further tumours in unsampled quadrants. Systematic sampling of the whole gland should be considered in addition to an MRI-targeted biopsy to capture both MRI-detected and MRI-undetected disease. Patient summaryAfter radiotherapy, magnetic resonance imaging (MRI) is accurate for detecting recurrent prostate cancer, with missed cancer being smaller and of lower grade. Targeting a biopsy to suspicious areas on MRI results in a diagnosis of cancer in most patients. However, for every five men who have recurrent cancer, this targeted approach would miss cancers elsewhere in the prostate in three of these men. If further focal treatment of the prostate is planned, random biopsies covering the whole prostate in addition to targeted biopsies should be considered so that tumours are not missed.

864 Does the PROMIS Trial Deliver What It Promised? Validation in a North West Population

Abstract Aim The PROMIS trial changed the landscape on how men with raised PSA are evaluated. We changed the pathway at a North West hospital, utilising multiparametric MRI (MP-MRI) scans and incorporated additional safety parameters (family history and PSA density) to maintain the advantage of avoiding unnecessary biopsy whilst not missing significant cancers. We present the performance of this pathway. Method We retrospectively reviewed 596 referrals of men with high PSA level between January - December 2021. Variables audited included demographics, PSA level, PSA density, Prostate Imaging-Reporting and Data System (PI-RADS) score on MP-MRI, biopsy results, Gleason score and prognostic grade grouping. Results Of 596 patients, 65.9% of men were aged between 50-75 years. The most common quintile of presenting PSA was 4.00-9.99 (53.2%). Of the 381 patients with non-metastatic prostate cancer who underwent MP-MRI, 38.3% were classed as PI-RAD 1-2, 16% PI-RAD 3 and 45.7% PI-RAD 4-5. Overall, 48.3% of patients did not undergo biopsy. Biopsy rates in PI-RAD 1-5 were 2.3%, 3.9%, 52.5%, 92.5% and 91.5% respectively, the decision made to monitor PSA instead in some cases due to co-morbidities or patient preference. Of 197 biopsies, 79.2% were positive, 87.8% initially classed as PI-RADS 4 or 5. Of the cancers picked up, 82.1% were clinically significant (Gleason score ≥3+4). Conclusions This study shows the pathway incorporating MP-MRI, family history and PSA density enables 48.3% of patients to avoid unnecessary biopsy and therefore overdiagnosis of clinically insignificant cancers. The PROMIS trial delivers what it promised.

Is Likert better than PI-RADS at predicting prostate cancer on MRI and can a mathematical algorithm achieve similar results?

Prostate Imaging Reporting & Data System (PI-RADS) is an internationally recognized system to quantify risk of prostate cancer on magnetic resonance imaging (MRI). However, studies have suggested methods to improve predictive accuracy. To assess two different methods that aim to improve the accuracy of PI-RADS scores: a subjective Likert score given by experienced reporters, and an objective Calculated Adjustment of PI-RADS Equivocal Score (CAPES). Five experienced reporters in a quaternary referral unit used a standardized reporting template to prospectively collect PI-RADS and Likert scores for 1467 multiparametric MRI (mpMRI) scans between January 2021 and June 2022. Histology results were recorded for patients who underwent trans-perineal biopsy. The CAPES tool was retrospectively applied to the cases scoring PI-RADS 3. A theoretical standardized biopsy protocol (assuming all patients scoring ≥3 were referred for biopsy) was used to compare the three scoring systems for sensitivity, specificity, and positive predictive value (PPV). Across all reporters, significantly fewer equivocal "3" scores were given using Likert (15.7%) or CAPES (2.2%) compared to PI-RADS (24.1%). Assuming a protocol where all patients scoring ≥3 were biopsied, Likert had a higher specificity (69.0% vs. 54.4%), sensitivity (98.3% vs. 97.7%), and PPV (49.9% vs. 40.3%) than PI-RADS for identifying ISUP ≥2 cancer. The CAPES tool had an even higher specificity (81.4%) and PPV (61.2%) with only a slightly lower sensitivity (93.4%) resulting in 37.1% (n = 316) fewer biopsies than PI-RADS, and 22.4% (n = 155) fewer biopsies than Likert across 1467 patients. Compared to PI-RADS scoring, Likert scoring or CAPES can result in fewer equivocal scores, greater PPV, and fewer unnecessary biopsies.

Comparison of model-free Lorentzian and spinlock model-based fittings in quantitative CEST imaging ofacutestroke.

CEST MRI detects complex tissue changes following acute stroke. Our study aimed to test if spinlock model-based fitting of the quasi-steady-state (QUASS)-reconstructed equilibrium CEST MRI improves the determination of multi-pool signal changes over the commonly-used model-free Lorentzian fitting in acute stroke. Multiple three-pool CEST Z-spectra were simulated using Bloch-McConnell equations for a range of T1 , relaxation delay, and saturation times. The multi-pool CEST signals were solved from the simulated Z-spectra to test the accuracy of routine Lorentzian (model-free) and spinlock (model-based) fittings without and with QUASS reconstruction. In addition, multiparametric MRI scans were obtained in rat models of acute stroke, including relaxation, diffusion, and CEST Z-spectrum. Finally, we compared model-free and model-based per-pixel CEST quantification in vivo. The spinlock model-based fitting of QUASS CEST MRI provided a nearly T1 -independent determination of multi-pool CEST signals, advantageous over the fittings of apparent CEST MRI (model-free and model-based). In vivo data also demonstrated that the spinlock model-based QUASS fitting captured significantly different changes in semisolid magnetization transfer (-0.9 ± 0.8 vs. 0.3 ± 0.8%), amide (-1.1 ± 0.4 vs. -0.5 ± 0.2%), and guanidyl (1.0 ± 0.4 vs. 0.7 ± 0.3%) signals over the model-free Lorentzian analysis. Our study demonstrated that spinlock model-based fitting of QUASS CEST MRI improved the determination of the underlying tissue changes following acute stroke, promising further clinical translation of quantitative CEST imaging.

IMG-07. MOLECULAR PROCESSES GIVE RISE TO EMERGENCE OF RADIO-PHENOTYPES ON MRI SCANS OF PEDIATRIC LOW-GRADE GLIOMA

Abstract There is mounting evidence that tumor microenvironmental pressure selects for somatic genetic alterations that contribute to the formation of distinct morphological characteristics, captured on radiology scans as radio-phenotypes. Such phenotypic variations are a source of heterogeneity in clinical manifestation of tumors of the same histology across the patients, and in part, their heterogeneous responses to therapies. Deeper understanding of the associations between genotype and radio-phenotype in pediatric low-grade glioma (pLGG), the most histologically diverse childhood brain tumor, may facilitate precision diagnostic and therapeutic approaches. Here, we categorize pLGGs into distinct and relatively homogeneous imaging subtypes based on radiomic features and further explore the associations of these imaging subtypes with genotype. From multiparametric MRI scans of 167 pLGGs from the Children’s Brain Tumor Network (CBTN), 881 radiomic features and clinical variables (tumor location, age, and gender) were extracted. After dimensionality reduction using principal component analysis (PCA), K-Means clustering was applied on 19 principal components to group the patients into three imaging subtypes. Using whole transcriptomic data from OpenTargets, differential expression and co-expression of network- and pathway-level and immune-related signaling were compared among these three imaging subtypes. Gene Set Enrichment Analysis (GSEA) revealed differentially higher expression of cell cycle regulatory, extracellular matrix (ECM) remodeling, and cell migratory pathways in imaging subtype1 than subtypes 2 and 3, and upregulation of ECM and immune-related pathways in subtypes 1 and 2 compared to subtype3. Based on Gene Sets Net Correlations Analysis (GSNCA), subtype1 exhibited differential co-regulation of TNF/TNFR1 signaling compared to subtype2, and differential co-regulation of RHOG GTPase and TGFB1pathways when compared against subtype3. Subtype2 showed differential co-regulation in NOTCH1 signaling and transcriptional regulatory pathways. Our proposed multi-disciplinary radiomic-genomic analysis approach elucidates the molecular and biological processes in the genotype of the tumors that are associated with emergence of distinct imaging subtypes in pLGG.

Prostate Imaging after Focal Ablation (PI-FAB): A Proposal for a Scoring System for Multiparametric MRI of the Prostate After Focal Therapy

At present there is no standardised system for scoring the appearance of the prostate on multiparametric magnetic resonance imaging (MRI) after focal ablation for localised prostate cancer. We propose a novel scoring system, the Prostate Imaging after Focal Ablation (PI-FAB) score, to fill this gap. PI-FAB involves a 3-point scale for rating MRI sequences in sequential order: (1) dynamic contrast-enhanced sequences; (2) diffusion-weighted imaging, split into assessment of the high-b-value sequence first and then the apparent diffusion coefficient map; and (3) T2-weighted imaging. It is essential that the pretreatment scan is also available to help with this assessment. We designed PI-FAB using our experience of reading postablation scans over the past 15 years and include details for four representative patients initially treated with high-intensity focus ultrasound at our institution to demonstrate the scoring system. We propose PI-FAB as a standardised method for evaluating prostate MRI scans after treatment with focal ablation. The next step is to evaluate its performance across multiple experienced readers of MRI after focal therapy in a clinical data set. Patient summaryWe propose a scoring system called PI-FAB for assessing the appearance of magnetic resonance imaging scans of the prostate after focal treatment for localised prostate cancer. This will help clinicians in deciding on further follow-up.

Prostate cancer screening: Have we tipped the seesaw?

43 Background: Screening for prostate cancer has been and remains contentious with evidence of harms and benefits finely balanced and inadequate to tip the see saw one way or another in order to give men clear, understandable guidance on PSA testing. In recent years the diagnostic pathway has changed significantly delivering a modern pathway that is likely to do far less harm and possibly more benefit. Our work quantifies the levels of harm and benefit at each stage of the entire pathway currently used in the UK from initial PSA through to biopsy-confirmed diagnosis of prostate cancer. Methods: Combining the full results of the most recent high quality diagnosis clinical trials, as well as real world evidence and practice patterns we have mapped each stage of the UK diagnostic pathway. Through this research we are able to quantify the number of men benefitting (i.e. those who move through the pathway correctly with an eventual diagnosis of clinically significant prostate cancer) and the number of men suffering harm (those who suffer biopsy side effects or who are subjected to unnecessary testing as they move through the pathway with no clinically significant cancer). Results: Initial results using clinical trial data and current practice patterns in the UK indicate that 10,000 men entering the diagnostic pathway with a PSA test would generate the following results. Diagnostic outcomes: 463 men would be diagnosed with a clinically significant (Gleason 3+4 or above) localised prostate cancer; 110 men would be diagnosed with a clinically insignificant (Gleason 3+3) prostate cancer; and 646 men would be taken through the entire diagnostic pathway with no diagnosis of prostate cancer. Harm: a maximum of 38 men would suffer clinical side effects from their biopsy. Healthcare costs: 10,000 PSA tests, 1,449 multiparametric MRI scans, and 1,207 biopsies would yield 463 significant cancers diagnosed. An estimated 1,258 clinically insignificant and 168 clinically significant cancers would remain undiagnosed using the current diagnostic pathway due to initial PSA below 3ng/ml. Conclusions: Rates of biopsy-related harm in modern practice are encouragingly low and likely to reduce further. A PSA-initiated diagnostic pathway misses a significant proportion of clinically important cancers. It is interesting to note the high conversion from MRI to biopsy in the data underpinning this model. Next steps will include analysis of the same outcomes in several centres in the UK to determine if actual practice matches expected outcomes (harms and benefits) from previous clinical trials.

Automated tumor segmentation and brain tissue extraction from multiparametric MRI of pediatric brain tumors: A multi-institutional study.

Brain tumors are the most common solid tumors and the leading cause of cancer-related death among all childhood cancers. Tumor segmentation is essential in surgical and treatment planning, and response assessment and monitoring. However, manual segmentation is time-consuming and has high interoperator variability. We present a multi-institutional deep learning-based method for automated brain extraction and segmentation of pediatric brain tumors based on multi-parametric MRI scans. Multi-parametric scans (T1w, T1w-CE, T2, and T2-FLAIR) of 244 pediatric patients ( n = 215 internal and n = 29 external cohorts) with de novo brain tumors, including a variety of tumor subtypes, were preprocessed and manually segmented to identify the brain tissue and tumor subregions into four tumor subregions, i.e., enhancing tumor (ET), non-enhancing tumor (NET), cystic components (CC), and peritumoral edema (ED). The internal cohort was split into training ( n = 151), validation ( n = 43), and withheld internal test ( n = 21) subsets. DeepMedic, a three-dimensional convolutional neural network, was trained and the model parameters were tuned. Finally, the network was evaluated on the withheld internal and external test cohorts. Dice similarity score (median ± SD) was 0.91 ± 0.10/0.88 ± 0.16 for the whole tumor, 0.73 ± 0.27/0.84 ± 0.29 for ET, 0.79 ± 19/0.74 ± 0.27 for union of all non-enhancing components (i.e., NET, CC, ED), and 0.98 ± 0.02 for brain tissue in both internal/external test sets. Our proposed automated brain extraction and tumor subregion segmentation models demonstrated accurate performance on segmentation of the brain tissue and whole tumor regions in pediatric brain tumors and can facilitate detection of abnormal regions for further clinical measurements.

Generalization of quasi-steady-state reconstruction to CEST MRI with two-tiered RF saturation and gradient-echo readout-Synergistic nuclear Overhauser enhancement contribution to brain tumor amide proton transfer-weighted MRI.

While amide proton transfer-weighted (APTw) MRI has been adopted in tumor imaging, there are concurrent APT, magnetization transfer, and nuclear Overhauser enhancement changes. Also, the APTw image is confounded by relaxation changes, particularly when the relaxation delay and saturation time are not sufficiently long. Our study aimed to extend a quasi-steady-state (QUASS) solution to determine the contribution of the multipool CEST signals to the observed tumor APTw contrast. Our study derived the QUASS solution for a multislice CEST-MRI sequence with an interleaved RF saturation and gradient-echo readout between signal averaging. Multiparametric MRI scans were obtained in rat brain tumor models, including T1 , T2 , diffusion, and CEST scans. Finally, we performed spinlock model-based multipool fitting to determine multiple concurrent CEST signal changes in the tumor. The QUASS APTw MRI showed small but significant differences in normal and tumor tissues and their contrast from the acquired asymmetry calculation. The spinlock model-based fitting showed significant differences in semisolid magnetization transfer, amide, and nuclear Overhauser enhancement effects between the apparent and QUASS CEST MRI. In addition, we determined that the tumor APTw contrast is due to synergistic APT increase (+3.5ppm) and NOE decrease (-3.5ppm), with their relative contribution being about one third and two thirds under a moderate B1 of 0.75 μT at 4.7 T. Our study generalized QUASS analysis to gradient-echo image readout and quantified the underlying tumor CEST signal changes, providing an improved elucidation of the commonly used APTw MRI.

Radiomic Analysis of Multi-parametric MR Images (MRI) for Classification of Parotid Tumors.

Characterization of parotid tumors before surgery using multi-parametric magnetic resonance imaging (MRI) scans can support clinical decision making about the best-suited therapeutic strategy for each patient. This study aims to differentiate benign from malignant parotid tumors through radiomics analysis of multi-parametric MR images, incorporating T2-w images with ADC-map and parametric maps generated from Dynamic Contrast Enhanced MRI (DCE-MRI). MRI scans of 31 patients with histopathologically-confirmed parotid gland tumors (23 benign, 8 malignant) were included in this retrospective study. For DCE-MRI, semi-quantitative analysis, Tofts pharmacokinetic (PK) modeling, and five-parameter sigmoid modeling were performed and parametric maps were generated. For each patient, borders of the tumors were delineated on whole tumor slices of T2-w image, ADC-map, and the late-enhancement dynamic series of DCE-MRI, creating regions-of-interest (ROIs). Radiomic analysis was performed for the specified ROIs. Among the DCE-MRI-derived parametric maps, wash-in rate (WIR) and PK-derived Ktrans parameters surpassed the accuracy of other parameters based on support vector machine (SVM) classifier. Radiomics analysis of ADC-map outperformed the T2-w and DCE-MRI techniques using the simpler classifier, suggestive of its inherently high sensitivity and specificity. Radiomics analysis of the combination of T2-w image, ADC-map, and DCE-MRI parametric maps resulted in accuracy of 100% with both classifiers with fewer numbers of selected texture features than individual images. In conclusion, radiomics analysis is a reliable quantitative approach for discrimination of parotid tumors and can be employed as a computer-aided approach for pre-operative diagnosis and treatment planning of the patients.

BIOM-63. IDENTIFICATION OF MIRNA IN CEREBROSPINAL FLUID AND PLASMA AS A BIOMARKER TO SUPPORT MRI EVALUATION AND MONITORING OF PEDIATRIC BRAIN TUMORS

Abstract MRI is the current gold standard imaging technique for diagnostic evaluation and monitoring of pediatric CNS tumors, however MRI measures are unable to fully relate to tumor biology and molecular stratification. Circulating in blood and cerebrospinal fluid (CSF), miRNAs are an abundant and stable nucleic acid which can be utilized as a tumor biomarker. Relating miRNA biomarkers and radiological tumor measurements may provide improved diagnostic and monitoring tools for pediatric brain tumors. Using a cohort of 54 pediatric brain tumors including low grade glioma, ependymoma, germ cell tumor, medulloblastoma, atypical teratoid rhabdoid tumor and high-grade glioma we attempted to combine MRI findings and circulating miRNA data. The miRNA expression was profiled in 33 CSF and 52 plasma samples using the HTG EdgeSeq platform. Clinically acquired, multi-parametric MRI scans at time-points close in proximity to liquid biopsy collection were collected retrospectively and used to generate volumetric tumor segmentations. We identified unique miRNA targets significantly correlated with MRI features, clinical findings, and patient outcomes. In both CSF and plasma, miRNA expression was identified to correlate with diagnosis and clinical features including tumor grade and survival status (p &amp;lt; 0.05). In CSF, miRNA expression was correlated with MRI measurements including cystic core volume, non-enhancing tumor volume, leptomeningeal disease, tumor size and location (p &amp;lt; 0.05). Combination of miRNA targets and radiomic tumor measurements improved diagnostic predictions between low- and high-grade tumors. In plasma, miRNA expression was correlated with MRI measurements including cystic core volume, location, and leptomeningeal disease (p &amp;lt; 0.05). These results demonstrate utility of miRNAs as a pediatric brain tumor biomarker which combined with imaging features can improve minimally to non-invasive diagnostics and management of pediatric brain tumors.

NIMG-37. JOINT LEARNING OF IMAGING AND GENOMIC DATA REVEALS DISTINCT GLIOBLASTOMA SUBTYPES

Abstract PURPOSE The significant heterogeneity of glioblastoma is typically displayed on both phenotypical and molecular levels. Non-invasive in vivo approaches to characterize this heterogeneity would potentially facilitate personalized therapies. Here we leverage advanced unsupervised machine learning to integrate radiomic imaging features and genomics to identify distinct subtypes of glioblastoma. METHODS A retrospective cohort of 571 IDH-wildtype glioblastoma patients were collected with pre-operative multi-parametric MRI (T1, T1CE, T2, T2-FLAIR, DSC, DTI) scans (available in 462/571 patients) and targeted next-generation sequencing (NGS) data (available in 355/571 patients). Radiomic features (n= 971) were extracted from these MRI scans and a subset of 12 features were selected by L21-norm minimization. A total of 14 key driver genes in the 5 main pathways that are most frequently altered in glioblastoma were chosen. Subtypes were identified by a joint learning approach called Anchor-based Partial Multi-modal Clustering (APMC) on both radiomic and genomic modalities. RESULTS Three distinct glioblastoma subtypes were discovered by APMC based on 14-dimension NGS data together with 12 selected radiomic features representing characteristics from histograms, shape, and volumetric measures for different tumor sub-regions. The identified subtypes were 1) high-risk; 2) medium-risk; and 3) low-risk, in terms of their overall survival outcome in Kaplan-Meier analysis (p= 5.52e-6, log-rank test; HR= 1.51, 95%CI:1.20-1.74, Cox proportional hazard model). The three subtypes also displayed different molecular characteristics: subtype 1 exhibited increased frequency of mutation in [EGFR, PIK3CA, PTEN, NF1], subtype 3 showed frequently mutated [PDGFRA, ATRX], while subtype 2 did not show significant differences for mutations in any of these genes. CONCLUSION Our results revealed the synergistic value of integrated radiomic signatures and molecular characteristics for glioblastoma subtyping. Joint learning on both modalities could help better understand the molecular basis of phenotypical signatures of glioblastoma and further provide insights into the biologic underpinnings of tumor formation and progression.

NIMG-82. RADIOGENOMIC SIGNATURES OF KEY DRIVER GENES IN GBM REVEAL MOLECULAR HETEROGENEITY OF THE TUMOR MICROENVIRONMENT LINKED TO SPATIAL DISTRIBUTION: IMPACT ON THE TRAJECTORY OF GLIOMA EVOLUTION

Abstract Somatic genomic alterations acquired during GBM growth enhance adaptation of tumor cells to their microenvironment and give rise to molecular heterogeneity. Radiogenomics could facilitate exploration of the underlying pathobiology of tumor growth in specific microenvironments and thereby, promote precision medicine for the patients. We derived radiogenomic signatures of key driver genes and evaluated molecular compositions of tumor groups with predisposition to specific brain regions. Pre-operative multiparametric conventional MRI scans of 357 IDH-wildtype GBM patients with available targeted NGS data were jointly segmented and registered into a common template. We constructed spatial distribution atlases for tumors harboring mutations in driver genes and identified four distinct groups of tumor locations with predilection to the left frontal cingulate region (Group1), right temporal (Group2), right parietal (Group3), and occipital pole (Group4). Evaluation of the differences in molecular features of the tumor groups included: (1) exploring similarities of genomic profiles across all four groups by evaluating cosine similarity metric (CSM) between mutational signatures; (2) quantification of molecular heterogeneity based on Mutant Allele Tumor Heterogeneity (MATH) scores; and (3) inference of the evolutionary trajectories. Groups 1 and 4 were the most different, and Groups 2 and 3 were the most similar tumors, molecularly. The mutational signatures between Groups 1 and 4 revealed a CSM of 0.35. Group1 showed significantly lower MATH score (less heterogeneity) compared to Group4 (p&amp;lt; 0.05). Evaluation of evolutionary patterns suggested NF1 mutation as an early event in Group1, without subsequent gain of function or mutation in EGFR. In contrast, in Group4, EGFR mutations were early events triggering PTEN mutations later in the evolutionary trajectory. Radiogenomic signatures revealed distinct molecular underpinnings for the tumors with predilection towards specific brain regions that may suggest existence of different tumor microenvironments in different brain regions that cause intra- and inter-patient heterogeneity in the molecular tumor composition.

NIMG-102. RAPNO-DEFINED SEGMENTATION AND VOLUMETRIC ASSESSMENT OF PEDIATRIC BRAIN TUMORS ON MULTI-PARAMETRIC MRI SCANS USING DEEP LEARNING; A ROBUST TOOL WITH POTENTIAL APPLICATION IN TUMOR RESPONSE ASSESSMENT

Abstract Volumetric measurements of whole tumor and its components on MRI scans, facilitated by automatic segmentation tools, are essential to reduce inter-observer variability in monitoring tumor progression and response assessment for pediatric brain tumors. Here, we present a fully automatic segmentation model based on deep learning that reliably delineates the tumor components recommended by the Response Assessment in Pediatric Neuro-Oncology (RAPNO) working group for evaluation of treatment response. Multi-parametric MRI (mpMRI) scans (T1-pre, T1-post, T2, and T2-FLAIR), acquired on multiple MRI scanners with different field strengths and vendors, for a cohort of 218 pediatric patients with a variety of histologically confirmed brain tumor subtypes were collected. The mpMRI scans were co-registered and manually segmented by experienced neuroradiologists in consensus to identify the tumor subregions including the enhancing tumor (ET), non-enhancing tumor (NET), cystic components (CC), and peritumoral edema (ED) regions. A convolutional neural network model based on DeepMedic architecture was trained using mpMRI scans as the inputs for segmentation of the whole tumor and subregions. The trained model showed excellent performance in segmentation of the whole tumor, as suggested by median dice of 0.90/0.85 for validation (n = 44)/independent test (n = 22) sets. ET and non-enhancing components (union of NET, CC, and ED) were segmented with median dice scores of 0.78/0.84 and 0.76/0.74 for validation/test sets, respectively. The automated and manual segmentations demonstrated strong agreement in estimating VASARI (Visually AcceSAble Rembrandt Images) MRI features with Pearson’s correlation coefficient R &amp;gt; 0.75 (p &amp;lt; 0.0001) for ET, NET, CC, and ED components. Our proposed automated segmentation method developed based on MRI scans acquired with different protocols, equipment, and from a variety of brain tumor subtypes, shows potential application for reliable and generalizable volumetric measurements which can be used for treatment response assessment in clinical trials.

NIMG-74. RADIOIMMUNOMIC SIGNATURES IN PEDIATRIC LOW-GRADE GLIOMA BASED ON MULTIPARAMETRIC MRI SCANS

Abstract Understanding the immune microenvironment in pediatric low-grade glioma (pLGG) patients may help in identification of the patients who benefit from anti-tumor immunotherapies. However, surgical resection is not feasible for many pLGG tumors in certain anatomical locations. Therefore, developing non-invasive tools that characterize the tumor microenvironment prior to therapeutic interventions could contribute to stratification and enrollment of the patients into relevant clinical trials. In this work, we derived radiomic signatures of immune profiles (radioimmunomics) based on machine learning (ML) analysis of readily available conventional MRI scans. Transcriptomic data for a cohort of 197 subjects was retrospectively collected from Open Pediatric Brain Tumor Atlas (OpenPBTA). The patients were categorized into three groups (Group1-3) based on their immunological profiles using consensus clustering algorithm. This analysis revealed greater immune cell infiltration in non-BRAF mutated pLGGs. Group1 showed more enrichment in M1 macrophages, and microenvironment and immune scores compared to Group2 and Group3. Elevated tumor inflammation score (TIS), as a predictor of clinical response to anti-PD-1 blockade, was observed in Group1 compared to Group2 (p= 1.4e-7) and Group3 (p= 0.0054). Radiomic features, including volumetric, morphologic, histogram, and texture descriptors, were extracted from the segmented tumor regions on multiparametric MRI (mpMRI) scans of 71 (of 197) patients. Multivariate ML models were trained to predict the three immunological groups based on radiomic features using cross-validated random forest classifier along with recursive feature elimination, which yielded AUC of 0.72 for this multi-class classification problem. Our findings indicate the presence of distinct immunological groups in pLGG tumors, with possibly more favorable response to immunotherapies in Group1 tumors. Furthermore, we developed radioimmunomic signatures based on pre-operative conventional mpMRI that can potentially stratify the patients based on their immune tumor microenvironment. Based on these initial promising results, we are exploring additional features to increase the accuracy of radioimmunomics model.

NIMG-25. OPTIMIZATION OF ARTIFICIAL INTELLIGENCE ALGORITHMS FOR LOW-RESOURCE/CLINICAL ENVIRONMENTS: FOCUS ON CLINICALLY-RELEVANT GLIOMA REGION DELINEATION

Abstract BACKGROUND Diffuse astrocytic glioma are common and aggressive malignant primary brain tumors with grim prognosis. Artificial intelligence (AI) has shown promise across predictive, prognostic, and diagnostic neuro-oncology applications, towards improving patient management. However, clinical translation and deployment are hampered by AI models’ requirements for explicit acceleration cards, which are not typically considered in clinical environments. Here, we seek the execution of AI models in such clinical/low-resource environments, facilitated by mathematical optimizations rather than investing in acceleration cards, and focus on the use case of delineating clinically-relevant regions, namely the entire tumor burden (ETB) and tumor core (TC). METHODS We identified the BraTS20 retrospective cohort of 369 glioma cases, each described by 4 structural multi-parametric MRI (mpMRI) scans (T1,T1Gd,T2,T2-FLAIR). The Generally Nuanced Deep Learning Framework (GaNDLFv0.0.14/PyTorchv1.8.2) was used to train the original AI model (ResUNet), by randomly sampling 40 643 patches from each mpMRI scan. We then investigate the contributions of post-training mathematical optimizations. Quantitative performance evaluation of the original and optimized (GaNDLFv0.0.14/OpenVINOTM-INT8-v2022.1.0) models were based on 125 unseen hold-out cases (BraTS20 validation dataset), using the dice similarity coefficient (DSC), while profiling in a consumer-grade workstation, i.e., typical clinical hardware configuration. RESULTS Negligible delineation performance differences were observed between the original and optimized AI models, for both ETB (DSCoriginal/DSCoptimized= 0.877/0.876) and TC (DSCoriginal/DSCoptimized= 0.773/0.772). However, the optimized model yielded substantial improvements in latency (up to 5.4x faster inference) and 53% less memory footprint. CONCLUSIONS Post-training mathematical optimization of AI models yields substantial gains in latency and memory usage, while maintaining their accuracy. Although we focused on delineation, we anticipate these mathematical optimizations to be applicable in other AI models. Post-training optimization is a promising approach for deploying AI models on consumer-grade CPUs, and hence facilitating their translation in low-resource/clinical environments, potentially contributing to improved patient management, treatment decisions, and response assessment. *author Siddhesh Thakur is an equal contributing first author.