Multicenter Dataset Research Articles

189P The synergic effect of metformin and atezolizumab/bevacizumab in diabetic HCC patients: A retrospective study from ARTE multicentric Italian dataset

189P The synergic effect of metformin and atezolizumab/bevacizumab in diabetic HCC patients: A retrospective study from ARTE multicentric Italian dataset

Delineating the Heterogeneity of Alzheimer’s Disease and Mild Cognitive Impairment Using Normative Models of Dynamic Brain Functional Networks

BackgroundAlzheimer’s disease (AD), which has been identified as the most common type of dementia, presents considerable heterogeneity in its clinical manifestations. Early intervention at the stage of mild cognitive impairment (MCI) holds potential in AD prevention. However, characterizing the heterogeneity of neurobiological abnormalities and identifying MCI subtypes pose significant challenges. MethodsWe constructed sex-specific normative age models of dynamic brain functional networks and mapped the deviations of the brain characteristics for individuals from multiple datasets, including 295 patients with AD, 441 patients with MCI, and 1160 normal control participants. Then, based on these individual deviation patterns, subtypes for both AD and MCI were identified using the clustering method, and their similarities and differences were comprehensively assessed. ResultsIndividuals with AD and MCI were clustered into 2 subtypes, and these subtypes exhibited significant differences in their intrinsic brain functional phenotypes and spatial atrophy patterns, as well as in disease progression and cognitive decline trajectories. The subtypes with positive deviations in AD and MCI shared similar deviation patterns, as did those with negative deviations. There was a potential transformation of MCI with negative deviation patterns into AD, and participants with MCI had a more severe cognitive decline rate. ConclusionsIn this study, we quantified neurophysiological heterogeneity by analyzing deviation patterns from the dynamic functional connectome normative model and identified disease subtypes of AD and MCI using a comprehensive resting-state functional magnetic resonance imaging multicenter dataset. The findings provide new insights for developing early prevention and personalized treatment strategies for AD.

No code machine learning: validating the approach on use-case for classifying clavicle fractures

PurposeWe created an infrastructure for no code machine learning (NML) platform for non-programming physicians to create NML model. We tested the platform by creating an NML model for classifying radiographs for the presence and absence of clavicle fractures. MethodsOur IRB-approved retrospective study included 4135 clavicle radiographs from 2039 patients (mean age 52 ± 20 years, F:M 1022:1017) from 13 hospitals. Each patient had two-view clavicle radiographs with axial and anterior-posterior projections. The positive radiographs had either displaced or non-displaced clavicle fractures. We configured the NML platform to automatically retrieve the eligible exams using the series' unique identification from the hospital virtual network archive via web access to DICOM Objects. The platform trained a model until the validation loss plateaus. Once the testing was complete, the platform provided the receiver operating characteristics curve and confusion matrix for estimating sensitivity, specificity, and accuracy. ResultsThe NML platform successfully retrieved 3917 radiographs (3917/4135, 94.7 %) and parsed them for creating a ML classifier with 2151 radiographs in the training, 100 radiographs for validation, and 1666 radiographs in testing datasets (772 radiographs with clavicle fracture, 894 without clavicle fracture). The network identified clavicle fracture with 90 % sensitivity, 87 % specificity, and 88 % accuracy with AUC of 0.95 (confidence interval 0.94–0.96). ConclusionA NML platform can help physicians create and test machine learning models from multicenter imaging datasets such as the one in our study for classifying radiographs based on the presence of clavicle fracture.

EXPLORING BACKDOOR ATTACKS IN OFF-THE-SHELF UNSUPERVISED DOMAIN ADAPTATION FOR SECURING CARDIAC MRI-BASED DIAGNOSIS.

The off-the-shelf model for unsupervised domain adaptation (OSUDA) has been introduced to protect patient data privacy and intellectual property of the source domain without access to the labeled source domain data. Yet, an off-the-shelf diagnosis model, deliberately compromised by backdoor attacks during the source domain training phase, can function as a parasite-host, disseminating the backdoor to the target domain model during the OSUDA stage. Because of limitations in accessing or controlling the source domain training data, OSUDA can make the target domain model highly vulnerable and susceptible to prominent attacks. To sidestep this issue, we propose to quantify the channel-wise backdoor sensitivity via a Lipschitz constant and, explicitly, eliminate the backdoor infection by overwriting the backdoor-related channel kernels with random initialization. Furthermore, we propose to employ an auxiliary model with a full source model to ensure accurate pseudo-labeling, taking into account the controllable, clean target training data in OSUDA. We validate our framework using a multi-center, multi-vendor, and multi-disease (M&M) cardiac dataset. Our findings suggest that the target model is susceptible to backdoor attacks during OSUDA, and our defense mechanism effectively mitigates the infection of target domain victims.

Surgical optomics: hyperspectral imaging and deep learning towards precision intraoperative automatic tissue recognition-results from the EX-MACHYNA trial.

Hyperspectral imaging (HSI), combined with machine learning, can help to identify characteristic tissue signatures enabling automatic tissue recognition during surgery. This study aims to develop the first HSI-based automatic abdominal tissue recognition with human data in a prospective bi-center setting. Data were collected from patients undergoing elective open abdominal surgery at two international tertiary referral hospitals from September 2020 to June 2021. HS images were captured at various time points throughout the surgical procedure. Resulting RGB images were annotated with 13 distinct organ labels. Convolutional Neural Networks (CNNs) were employed for the analysis, with both external and internal validation settings utilized. A total of 169 patients were included, 73 (43.2%) from Strasbourg and 96 (56.8%) from Verona. The internal validation within centers combined patients from both centers into a single cohort, randomly allocated to the training (127 patients, 75.1%, 585 images) and test sets (42 patients, 24.9%, 181 images). This validation setting showed the best performance. The highest true positive rate was achieved for the skin (100%) and the liver (97%). Misclassifications included tissues with a similar embryological origin (omentum and mesentery: 32%) or with overlaying boundaries (liver and hepatic ligament: 22%). The median DICE score for ten tissue classes exceeded 80%. To improve automatic surgical scene segmentation and to drive clinical translation, multicenter accurate HSI datasets are essential, but further work is needed to quantify the clinical value of HSI. HSI might be included in a new omics science, namely surgical optomics, which uses light to extract quantifiable tissue features during surgery.

#1254 SGLT-2 inhibitors in patients with type 2 diabetes with acute kidney disease

Abstract Background and Aims Sodium-glucose cotransporter-2 (SGLT-2) inhibitors have been observed to exhibit correlative relationships with favorable impacts on renal and cardiovascular outcomes among individuals diagnosed with type 2 diabetes. Nonetheless, the extent of the influence of SGLT-2 inhibitors on patients with type 2 diabetes concomitant with acute kidney disease (AKD) remains enigmatic. Method This investigation leveraged expansive global healthcare data from TriNetX, spanning the timeframe from September 2002 to September 2022. Propensity score matching was systematically applied to meticulously curate a patient cohort, with a subsequent follow-up period extending over 5 years or until the manifestation of a defined outcome or mortality. The underlying research design took the form of an observational cohort study. The primary objective of this study was to scrutinize the enduring associations of Sodium-glucose cotransporter-2 (SGLT-2) inhibitors in patients afflicted with type 2 diabetes and concurrent acute kidney disease (AKD). Particular emphasis was directed towards evaluating the impact on mortality rates, major adverse kidney events (MAKEs), and major adverse cardiovascular events (MACEs) over the specified temporal continuum. The primary outcomes measured were mortality, MAKEs, and MACEs. Adjusted hazard ratios (aHR) with 95% confidence intervals (CI) were calculated to compare the risks between SGLT-2 inhibitor users and non-users, representing the average treatment effect among the treated. Results A total cohort of 230, 366 individuals diagnosed with AKD, boasting a mean age of 67.1 years and a male representation of 53.1%, formed the basis of this study. Within this cohort, 5 319 subjects (constituting 2.3%) were identified as users of SGLT-2 inhibitors. The incidence rates for mortality, MAKEs, and MACEs among non-users were recorded at 18.7%, 21.0%, and 25.8%, respectively. Following the meticulous application of propensity score matching, absolute differences in the incidence rates of mortality, MAKEs, and MACEs between SGLT-2 inhibitor users and non-users were calculated at 9.7%, 11.5%, and 12.3%, respectively. In the treated population, SGLT-2 inhibitor users exhibited a notably lower risk of mortality (adjusted Hazard Ratio [aHR]: 0.69, 95% Confidence Interval [CI]: 0.62-0.77), MAKEs (aHR: 0.62, 95% CI: 0.56-0.69), and MACEs (aHR: 0.75, 95% CI: 0.65-0.88) in comparison to their non-using counterparts. The robustness of these observed benefits was validated externally using a multicenter cohort dataset comprising 1, 233 AKD patients who were SGLT-2 inhibitor users, thereby affirming the positive outcomes. Importantly, the risk reduction associated with SGLT-2 inhibitors persisted significantly even among subpopulations devoid of hypertension, those with advanced chronic kidney disease, and individuals not concurrently receiving other oral hypoglycemic agents. Conclusion In this meticulously conducted cohort study involving individuals diagnosed with type 2 diabetes and AKD receiving treatment with SGLT-2 inhibitors, our findings suggest a potential association with diminished rates of mortality, MAKEs, and MACEs. These results underscore the imperative for subsequent and more in-depth investigations into the precise associations of SGLT-2 inhibitors within this distinctively treated population.

HCA-DAN: hierarchical class-aware domain adaptive network for gastric tumor segmentation in 3D CT images

BackgroundAccurate segmentation of gastric tumors from CT scans provides useful image information for guiding the diagnosis and treatment of gastric cancer. However, automated gastric tumor segmentation from 3D CT images faces several challenges. The large variation of anisotropic spatial resolution limits the ability of 3D convolutional neural networks (CNNs) to learn features from different views. The background texture of gastric tumor is complex, and its size, shape and intensity distribution are highly variable, which makes it more difficult for deep learning methods to capture the boundary. In particular, while multi-center datasets increase sample size and representation ability, they suffer from inter-center heterogeneity.MethodsIn this study, we propose a new cross-center 3D tumor segmentation method named Hierarchical Class-Aware Domain Adaptive Network (HCA-DAN), which includes a new 3D neural network that efficiently bridges an Anisotropic neural network and a Transformer (AsTr) for extracting multi-scale context features from the CT images with anisotropic resolution, and a hierarchical class-aware domain alignment (HCADA) module for adaptively aligning multi-scale context features across two domains by integrating a class attention map with class-specific information. We evaluate the proposed method on an in-house CT image dataset collected from four medical centers and validate its segmentation performance in both in-center and cross-center test scenarios.ResultsOur baseline segmentation network (i.e., AsTr) achieves best results compared to other 3D segmentation models, with a mean dice similarity coefficient (DSC) of 59.26%, 55.97%, 48.83% and 67.28% in four in-center test tasks, and with a DSC of 56.42%, 55.94%, 46.54% and 60.62% in four cross-center test tasks. In addition, the proposed cross-center segmentation network (i.e., HCA-DAN) obtains excellent results compared to other unsupervised domain adaptation methods, with a DSC of 58.36%, 56.72%, 49.25%, and 62.20% in four cross-center test tasks.ConclusionsComprehensive experimental results demonstrate that the proposed method outperforms compared methods on this multi-center database and is promising for routine clinical workflows.

Multi-centre benchmarking of deep learning models for COVID-19 detection in chest x-rays.

This study is a retrospective evaluation of the performance of deep learning models that were developed for the detection of COVID-19 from chest x-rays, undertaken with the goal of assessing the suitability of such systems as clinical decision support tools. Models were trained on the National COVID-19 Chest Imaging Database (NCCID), a UK-wide multi-centre dataset from 26 different NHS hospitals and evaluated on independent multi-national clinical datasets. The evaluation considers clinical and technical contributors to model error and potential model bias. Model predictions are examined for spurious feature correlations using techniques for explainable prediction. Models performed adequately on NHS populations, with performance comparable to radiologists, but generalised poorly to international populations. Models performed better in males than females, and performance varied across age groups. Alarmingly, models routinely failed when applied to complex clinical cases with confounding pathologies and when applied to radiologist defined "mild" cases. This comprehensive benchmarking study examines the pitfalls in current practices that have led to impractical model development. Key findings highlight the need for clinician involvement at all stages of model development, from data curation and label definition, to model evaluation, to ensure that all clinical factors and disease features are appropriately considered during model design. This is imperative to ensure automated approaches developed for disease detection are fit-for-purpose in a clinical setting.

Challenges in multi-centric generalization: phase and step recognition in Roux-en-Y gastric bypass surgery

PurposeMost studies on surgical activity recognition utilizing artificial intelligence (AI) have focused mainly on recognizing one type of activity from small and mono-centric surgical video datasets. It remains speculative whether those models would generalize to other centers.MethodsIn this work, we introduce a large multi-centric multi-activity dataset consisting of 140 surgical videos (MultiBypass140) of laparoscopic Roux-en-Y gastric bypass (LRYGB) surgeries performed at two medical centers, i.e., the University Hospital of Strasbourg, France (StrasBypass70) and Inselspital, Bern University Hospital, Switzerland (BernBypass70). The dataset has been fully annotated with phases and steps by two board-certified surgeons. Furthermore, we assess the generalizability and benchmark different deep learning models for the task of phase and step recognition in 7 experimental studies: (1) Training and evaluation on BernBypass70; (2) Training and evaluation on StrasBypass70; (3) Training and evaluation on the joint MultiBypass140 dataset; (4) Training on BernBypass70, evaluation on StrasBypass70; (5) Training on StrasBypass70, evaluation on BernBypass70; Training on MultiBypass140, (6) evaluation on BernBypass70 and (7) evaluation on StrasBypass70.ResultsThe model’s performance is markedly influenced by the training data. The worst results were obtained in experiments (4) and (5) confirming the limited generalization capabilities of models trained on mono-centric data. The use of multi-centric training data, experiments (6) and (7), improves the generalization capabilities of the models, bringing them beyond the level of independent mono-centric training and validation (experiments (1) and (2)).ConclusionMultiBypass140 shows considerable variation in surgical technique and workflow of LRYGB procedures between centers. Therefore, generalization experiments demonstrate a remarkable difference in model performance. These results highlight the importance of multi-centric datasets for AI model generalization to account for variance in surgical technique and workflows. The dataset and code are publicly available at https://github.com/CAMMA-public/MultiBypass140.

Advancing Automatic Gastritis Diagnosis: An Interpretable Multilabel Deep Learning Framework for the Simultaneous Assessment of Multiple Indicators

The evaluation of morphological features such as inflammation, gastric atrophy, and intestinal metaplasia is crucial for diagnosing gastritis. However, artificial intelligence (AI) analysis for nontumor diseases like gastritis is limited. Previous deep learning models have omitted important morphological indicators and cannot simultaneously diagnose gastritis indicators or provide interpretable labels. To address this, an attention-based multi-instance multilabel learning network (AMMNet) was developed to simultaneously achieve the multilabel diagnosis of activity, atrophy, and intestinal metaplasia with only slide-level weak labels. To evaluate AMMNet's real-world performance, a diagnostic test was designed to observe improvements in junior pathologists' diagnostic accuracy and efficiency with and without AMMNet assistance. In this study of 1,096 patients from 7 independent medical centers, AMMNet performed well in assessing activity (area under the curve (AUC): 0.93), atrophy (AUC: 0.97), and intestinal metaplasia (AUC: 0.93). The false-negative rates (FNRs) of these indicators were only 0.04, 0.08, and 0.18, respectively, and junior pathologists had lower FNRs with model assistance (0.15 vs. 0.10). Furthermore, AMMNet reduced the time required per whole-slide image (WSI) from 5.46 minutes to only 2.85 minutes, enhancing diagnostic efficiency. In block-level clustering analysis, AMMNet effectively visualized task-related patches within WSIs, improving interpretability. These findings highlight AMMNet’s effectiveness in accurately evaluating gastritis morphological indicators on multicenter datasets. Using multi-instance multilabel learning strategies to support routine diagnostic pathology deserves further evaluation.

Breast Delineation in Full-Field Digital Mammography Using the Segment Anything Model

Breast cancer is a major health concern worldwide. Mammography, a cost-effective and accurate tool, is crucial in combating this issue. However, low contrast, noise, and artifacts can limit the diagnostic capabilities of radiologists. Computer-Aided Diagnosis (CAD) systems have been developed to overcome these challenges, with the accurate outlining of the breast being a critical step for further analysis. This study introduces the SAM-breast model, an adaptation of the Segment Anything Model (SAM) for segmenting the breast region in mammograms. This method enhances the delineation of the breast and the exclusion of the pectoral muscle in both medio lateral-oblique (MLO) and cranio-caudal (CC) views. We trained the models using a large, multi-center proprietary dataset of 2492 mammograms. The proposed SAM-breast model achieved the highest overall Dice Similarity Coefficient (DSC) of 99.22% ± 1.13 and Intersection over Union (IoU) 98.48% ± 2.10 over independent test images from five different datasets (two proprietary and three publicly available). The results are consistent across the different datasets, regardless of the vendor or image resolution. Compared with other baseline and deep learning-based methods, the proposed method exhibits enhanced performance. The SAM-breast model demonstrates the power of the SAM to adapt when it is tailored to specific tasks, in this case, the delineation of the breast in mammograms. Comprehensive evaluations across diverse datasets—both private and public—attest to the method’s robustness, flexibility, and generalization capabilities.

Prediction of heart failure patients with distinct left ventricular ejection fraction levels using circadian ECG features and machine learning.

Heart failure (HF) encompasses a diverse clinical spectrum, including instances of transient HF or HF with recovered ejection fraction, alongside persistent cases. This dynamic condition exhibits a growing prevalence and entails substantial healthcare expenditures, with anticipated escalation in the future. It is essential to classify HF patients into three groups based on their ejection fraction: reduced (HFrEF), mid-range (HFmEF), and preserved (HFpEF), such as for diagnosis, risk assessment, treatment choice, and the ongoing monitoring of heart failure. Nevertheless, obtaining a definitive prediction poses challenges, requiring the reliance on echocardiography. On the contrary, an electrocardiogram (ECG) provides a straightforward, quick, continuous assessment of the patient's cardiac rhythm, serving as a cost-effective adjunct to echocardiography. In this research, we evaluate several machine learning (ML)-based classification models, such as K-nearest neighbors (KNN), neural networks (NN), support vector machines (SVM), and decision trees (TREE), to classify left ventricular ejection fraction (LVEF) for three categories of HF patients at hourly intervals, using 24-hour ECG recordings. Information from heterogeneous group of 303 heart failure patients, encompassing HFpEF, HFmEF, or HFrEF classes, was acquired from a multicenter dataset involving both American and Greek populations. Features extracted from ECG data were employed to train the aforementioned ML classification models, with the training occurring in one-hour intervals. To optimize the classification of LVEF levels in coronary artery disease (CAD) patients, a nested cross-validation approach was employed for hyperparameter tuning. HF patients were best classified using TREE and KNN models, with an overall accuracy of 91.2% and 90.9%, and average area under the curve of the receiver operating characteristics (AUROC) of 0.98, and 0.99, respectively. Furthermore, according to the experimental findings, the time periods of midnight-1 am, 8-9 am, and 10-11 pm were the ones that contributed to the highest classification accuracy. The results pave the way for creating an automated screening system tailored for patients with CAD, utilizing optimal measurement timings aligned with their circadian cycles.

Reproducible Radiomics Features from Multi-MRI-Scanner Test-Retest-Study: Influence on Performance and Generalizability of Models.

Radiomics models trained on data from one center typically show a decline of performance when applied to data from external centers, hindering their introduction into large-scale clinical practice. Current expert recommendations suggest to use only reproducible radiomics features isolated by multiscanner test-retest experiments, which might help to overcome the problem of limited generalizability to external data. To evaluate the influence of using only a subset of robust radiomics features, defined in a prior in vivo multi-MRI-scanner test-retest-study, on the performance and generalizability of radiomics models. Retrospective. Patients with monoclonal plasma cell disorders. Training set (117 MRIs from center 1); internal test set (42 MRIs from center 1); external test set (143 MRIs from center 2-8). 1.5T and 3.0T; T1-weighted turbo spin echo. The task for the radiomics models was to predict plasma cell infiltration, determined by bone marrow biopsy, noninvasively from MRI. Radiomics machine learning models, including linear regressor, support vector regressor (SVR), and random forest regressor (RFR), were trained on data from center 1, using either all radiomics features, or using only reproducible radiomics features. Models were tested on an internal (center 1) and a multicentric external data set (center 2-8). Pearson correlation coefficient r and mean absolute error (MAE) between predicted and actual plasma cell infiltration. Fisher's z-transformation, Wilcoxon signed-rank test, Wilcoxon rank-sum test; significance level P < 0.05. When using only reproducible features compared with all features, the performance of the SVR on the external test set significantly improved (r = 0.43 vs. r = 0.18 and MAE = 22.6 vs. MAE = 28.2). For the RFR, the performance on the external test set deteriorated when using only reproducible instead of all radiomics features (r = 0.33 vs. r = 0.44, P = 0.29 and MAE = 21.9 vs. MAE = 20.5, P = 0.10). Using only reproducible radiomics features improves the external performance of some, but not all machine learning models, and did not automatically lead to an improvement of the external performance of the overall best radiomics model. Stage 2.

Classification of lung cancer subtypes on CT images with synthetic pathological priors

The accurate diagnosis on pathological subtypes for lung cancer is of significant importance for the follow-up treatments and prognosis managements. In this paper, we propose self-generating hybrid feature network (SGHF-Net) for accurately classifying lung cancer subtypes on computed tomography (CT) images. Inspired by studies stating that cross-scale associations exist in the image patterns between the same case’s CT images and its pathological images, we innovatively developed a pathological feature synthetic module (PFSM), which quantitatively maps cross-modality associations through deep neural networks, to derive the “gold standard” information contained in the corresponding pathological images from CT images. Additionally, we designed a radiological feature extraction module (RFEM) to directly acquire CT image information and integrated it with the pathological priors under an effective feature fusion framework, enabling the entire classification model to generate more indicative and specific pathologically related features and eventually output more accurate predictions. The superiority of the proposed model lies in its ability to self-generate hybrid features that contain multi-modality image information based on a single-modality input. To evaluate the effectiveness, adaptability, and generalization ability of our model, we performed extensive experiments on a large-scale multi-center dataset (i.e., 829 cases from three hospitals) to compare our model and a series of state-of-the-art (SOTA) classification models. The experimental results demonstrated the superiority of our model for lung cancer subtypes classification with significant accuracy improvements in terms of accuracy (ACC), area under the curve (AUC), positive predictive value (PPV) and F1-score.

Current Diagnosis and Treatment of Acute Pancreatitis in Korea: A Nationwide Survey.

Acute pancreatitis (AP) is a leading cause of emergency hospitalization. We present the current diagnostic and therapeutic status of AP as revealed by analysis of a large multicenter dataset. The medical records of patients diagnosed with AP between 2018 and 2019 in 12 tertiary medical centers in Korea were retrospectively reviewed. In total, 676 patients were included, of whom 388 (57.4%) were male, and the mean age of all patients was 58.6 years. There were 355 (52.5%), 301 (44.5%), and 20 (3.0%) patients with mild, moderate, and severe AP, respectively, as assessed by the revised Atlanta classification. The most common etiologies of AP were biliary issues (41.6%) and alcohol consumption (24.6%), followed by hypertriglyceridemia (6.8%). The etiology was not identified in 111 (16.4%) patients at the time of initial admission. The overall mortality rate was 3.3%, increasing up to 45.0% among patients with severe AP. Notably, 70.0% (14/20) of patients with severe AP and 81.5% (154/189) of patients with systemic inflammatory response syndrome had received <4 L per day during the initial 24 hours of admission. Only 23.8% (67/281) of acute biliary pancreatitis patients underwent cholecystectomy during their initial admission. In total, 17.8% of patients experienced recurrent attacks during follow-up. However, none of the patients with acute biliary pancreatitis experienced recurrent attacks if they had undergone cholecystectomy during their initial admission. This study provides insights into the current status of AP in Korea, including its etiology, severity, and management. Results reveal disparities between clinical guidelines and their practical implementation for AP treatment.

T2-weighted imaging-based deep-learning method for noninvasive prostate cancer detection and Gleason grade prediction: a multicenter study.

To noninvasively detect prostate cancer and predict the Gleason grade using single-modality T2-weighted imaging with a deep-learning approach. Patients with prostate cancer, confirmed by histopathology, who underwent magnetic resonance imaging examinations at our hospital during September 2015-June 2022 were retrospectively included in an internal dataset. An external dataset from another medical center and a public challenge dataset were used for external validation. A deep-learning approach was designed for prostate cancer detection and Gleason grade prediction. The area under the curve (AUC) was calculated to compare the model performance. For prostate cancer detection, the internal datasets comprised data from 195 healthy individuals (age: 57.27 ± 14.45 years) and 302 patients (age: 72.20 ± 8.34 years) diagnosed with prostate cancer. The AUC of our model for prostate cancer detection in the validation set (n = 96, 19.7%) was 0.918. For Gleason grade prediction, datasets comprising data from 283 of 302 patients with prostate cancer were used, with 227 (age: 72.06 ± 7.98 years) and 56 (age: 72.78 ± 9.49 years) patients being used for training and testing, respectively. The external and public challenge datasets comprised data from 48 (age: 72.19 ± 7.81 years) and 91 patients (unavailable information on age), respectively. The AUC of our model for Gleason grade prediction in the training set (n = 227) was 0.902, whereas those of the validation (n = 56), external validation (n = 48), and public challenge validation sets (n = 91) were 0.854, 0.776, and 0.838, respectively. Through multicenter dataset validation, our proposed deep-learning method could detect prostate cancer and predict the Gleason grade better than human experts. Precise prostate cancer detection and Gleason grade prediction have great significance for clinical treatment and decision making. Prostate segmentation is easier to annotate than prostate cancer lesions for radiologists. Our deep-learning method detected prostate cancer and predicted the Gleason grade, outperforming human experts. Non-invasive Gleason grade prediction can reduce the number of unnecessary biopsies.

Abstract PO1-07-11: Radiogenomic-based imaging intratumor heterogeneity model predicts breast cancer prognosis and unveils therapeutic targets

Abstract Background: Intratumor heterogeneity (ITH) refers to variations observed among cancer cells within a single tumor, posing significant challenges in clinical practice due to its contribution to treatment resistance and unfavorable outcome. However, the inconvenience of multi-region biopsy and limitations of genomic sequencing based on limited tissue impede the practical detection of ITH. Consequently, there is an urgent need for a non-invasive method that comprehensively captures ITH from a holistic perspective of the entire tumor. Methods: We utilized a large multicenter dataset comprising dynamic contrast-enhanced magnetic resonance images from breast cancer patients (n = 1423) along with matched multiomics data (n = 468) to develop a non-invasive machine learning approach for measuring ITH. We extracted quantitative radiomic features from both the entire tumor and peritumor regions, with a specific focus on features associated with imaging heterogeneity. Using these radiomic features, we established an imaging ITH (IITH) model. The robustness of IITH evaluation was determined by assessing its correlation with genomic ITH (employing the mutant-allele tumor heterogeneity [MATH] algorithm) and pathological cellular ITH (analyzing variation in quantitative nuclear features extracted through digital pathology). Prognostic power was evaluated using Kaplan-Meier and multivariate Cox proportional hazards analyses. Integrated multiomics analyses were performed to investigate the molecular basis of different IITH subgroups. Results: We developed a non-invasive radiomic signature to quantify imaging ITH (IITH) in the FUSCC cohort. Breast cancer patients were retrospectively categorized into high and low IITH groups. Multivariate Cox analysis identified high IITH as an independent predictor of poor prognosis in breast cancer patients, even after adjusting for clinical risk factors such as tumor size, positive lymph nodes, clinical subtype and lymphovascular invasion status. These findings were further validated in the DUKE cohort, confirming the prognostic value of IITH. Moreover, we substantiated the robustness of IITH by demonstrating its association with genomic and pathological ITH. Multiomics analysis revealed the activation of oncogenic pathways and metabolic dysregulation in high-IITH tumors. Intriguingly, our investigation also highlighted ferroptosis as a vulnerability and potential therapeutic target in high-IITH tumors, which was further supported by evidence from the TCGA cohort. Conclusion: Radiomic-based assessment of ITH provides a non-invasive approach to comprehensively capture ITH and predict the prognosis of breast cancer patients. Targeting ferroptosis may hold promise as a treatment strategy for patients with high IITH. Citation Format: Guan-Hua Su, Yi Xiao, Chao You, Ren-Cheng Zheng, Shen Zhao, He Wang, Yi-Zhou Jiang, Ya-Jia Gu, Zhi-Ming Shao. Radiogenomic-based imaging intratumor heterogeneity model predicts breast cancer prognosis and unveils therapeutic targets [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO1-07-11.

CAVE: Cerebral artery–vein segmentation in digital subtraction angiography

Cerebral X-ray digital subtraction angiography (DSA) is a widely used imaging technique in patients with neurovascular disease, allowing for vessel and flow visualization with high spatio-temporal resolution. Automatic artery–vein segmentation in DSA plays a fundamental role in vascular analysis with quantitative biomarker extraction, facilitating a wide range of clinical applications. The widely adopted U-Net applied on static DSA frames often struggles with disentangling vessels from subtraction artifacts. Further, it falls short in effectively separating arteries and veins as it disregards the temporal perspectives inherent in DSA. To address these limitations, we propose to simultaneously leverage spatial vasculature and temporal cerebral flow characteristics to segment arteries and veins in DSA. The proposed network, coined CAVE, encodes a 2D+time DSA series using spatial modules, aggregates all the features using temporal modules, and decodes it into 2D segmentation maps. On a large multi-center clinical dataset, CAVE achieves a vessel segmentation Dice of 0.84 (±0.04) and an artery–vein segmentation Dice of 0.79 (±0.06). CAVE surpasses traditional Frangi-based k-means clustering (P < 0.001) and U-Net (P < 0.001) by a significant margin, demonstrating the advantages of harvesting spatio-temporal features. This study represents the first investigation into automatic artery–vein segmentation in DSA using deep learning. The code is publicly available at https://github.com/RuishengSu/CAVE_DSA.

Robust Ensemble of Two Different Multimodal Approaches to Segment 3D Ischemic Stroke Segmentation Using Brain Tumor Representation Among Multiple Center Datasets.

Ischemic stroke segmentation at an acute stage is vital in assessing the severity of patients' impairment and guiding therapeutic decision-making for reperfusion. Although many deep learning studies have shown attractive performance in medical segmentation, it is difficult to use these models trained on public data with private hospitals' datasets. Here, we demonstrate an ensemble model that employs two different multimodal approaches for generalization, a more effective way to perform on external datasets. First, after we jointly train a segmentation model on diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) MR modalities, the model is inferred on the DWI images. Second, a channel-wise segmentation model is trained by concatenating the DWI and ADC images as input, and then is inferred using both MR modalities. Before training with ischemic stroke data, we utilized BraTS 2021, a public brain tumor dataset, for transfer learning. An extensive ablation study evaluates which strategy learns better representations for ischemic stroke segmentation. In our study, nnU-Net well-known for robustness is selected as our baseline model. Our proposed method is evaluated on three different datasets: the Asan Medical Center (AMC) I and II, and the 2022 Ischemic Stroke Lesion Segmentation (ISLES). Our experiments are widely validated over a large, multi-center, and multi-scanner dataset with a huge amount of 846 scans. Not only stroke lesion models can benefit from transfer learning using brain tumor data, but combining the MR modalities using different training schemes also highly improves segmentation performance. The method achieved a top-1 ranking in the ongoing ISLES'22 challenge and performed particularly well on lesion-wise metrics of interest to neuroradiologists, achieving a Dice coefficient of 78.69% and a lesion-wise F1 score of 82.46%. Also, the method was relatively robust on the AMC I (Dice, 60.35%; lesion-wise F1, 68.30%) and II (Dice; 74.12%; lesion-wise F1, 67.53%) datasets in different settings. The high segmentation accuracy of our proposed method could improve radiologists' ability to detect ischemic stroke lesions in MRI images. Our model weights and inference code are available on https://github.com/MDOpx/ISLES22-model-inference .

Prediction models for postoperative renal function after living donor nephrectomy: a systematic review.

Living-donor nephrectomy (LDN) is the most valuable source of organs for kidney transplantation worldwide. The current preoperative evaluation of a potential living donor candidate does not take into account formal estimation of postoperative renal function decline after surgery using validated prediction models. The aim of this study was to summarize the available models to predict the mid- to long-term renal function following LDN, aiming to support both clinicians and patients during the decision-making process. A systematic review of the English-language literature was conducted following the principles highlighted by the European Association of Urology (EAU) guidelines and following the PRISMA 2020 recommendations. The protocol was registered in PROSPERO on December 10, 2022 (registration ID: CRD42022380198). In the qualitative analysis we selected the models including only preoperative variables. After screening and eligibility assessment, six models from six studies met the inclusion criteria. All of them relied on retrospective patient cohorts. According to PROBAST, all studies were evaluated as high risk of bias. The models included different combinations of variables (ranging between two to four), including donor-/kidney-related factors, and preoperative laboratory tests. Donor age was the variable more often included in the models (83%), followed by history of hypertension (17%), Body Mass Index (33%), renal volume adjusted by body weight (33%) and body surface area (33%). There was significant heterogeneity in the model building strategy, the main outcome measures and the model's performance metrics. Three models were externally validated. Few models using preoperative variables have been developed and externally validated to predict renal function after LDN. As such, the evidence is premature to recommend their use in routine clinical practice. Future research should be focused on the development and validation of user-friendly, robust prediction models, relying on granular large multicenter datasets, to support clinicians and patients during the decision-making process.