Advances in medical imaging and deep learning have propelled progress in whole slide image (WSI) analysis, with multiple instance learning (MIL) showing promise for efficient and accurate diagnostics. However, conventional MIL models often lack adaptability to evolving datasets, as they rely on static training that cannot incorporate new information without extensive retraining. Applying continual learning (CL) to MIL models is a possible solution, but often sees limited improvements. In this paper, we analyze CL in the context of attention MIL models and find that the model forgetting is mainly concentrated in the attention layers of the MIL model. Using the results of this analysis we propose two components for improving CL on MIL: Attention Knowledge Distillation (AKD) and the Pseudo-Bag Memory Pool (PMP). AKD mitigates catastrophic forgetting by focusing on retaining attention layer knowledge between learning sessions, while PMP reduces the memory footprint by selectively storing only the most informative patches, or ``pseudo-bags'' from WSIs. Experimental evaluations demonstrate that our method significantly improves both accuracy and memory efficiency on diverse WSI datasets, outperforming current state-of-the-art CL methods. This work provides a foundation for CL in large-scale, weakly annotated clinical datasets, paving the way for more adaptable and resilient diagnostic models.

Glioblastoma (GBM) is the most aggressive adult primary brain cancer, characterized by significant heterogeneity, posing challenges for patient management, treatment planning, and clinical trial stratification. We developed a highly reproducible, personalized prognostication, and clinical subgrouping system using machine learning (ML) on routine clinical data, magnetic resonance imaging (MRI), and molecular measures from 2838 demographically diverse patients across 22 institutions and 3 continents. Patients were stratified into favorable, intermediate, and poor prognostic subgroups (I, II, and III) using Kaplan-Meier analysis (Cox proportional model and hazard ratios [HR]). The ML model stratified patients into distinct prognostic subgroups with HRs between subgroups I-II and I-III of 1.62 (95% CI: 1.43-1.84, P < .001) and 3.48 (95% CI: 2.94-4.11, P < .001), respectively. Analysis of imaging features revealed several tumor properties contributing unique prognostic value, supporting the feasibility of a generalizable prognostic classification system in a diverse cohort. Our ML model demonstrates extensive reproducibility and online accessibility, utilizing routine imaging data rather than complex imaging protocols. This platform offers a unique approach to personalized patient management and clinical trial stratification in GBM.

Deep learning models hold significant promise for disease diagnosis but often lack transparency in their decision-making processes, limiting trust and hindering clinical adoption. This study introduces a novel multi-task learning framework to enhance the medical explainability of deep learning models for diagnosing age-related macular degeneration (AMD) using fundus images. The framework simultaneously performs AMD classification and lesion segmentation, allowing the model to support its diagnoses with AMD-associated lesions identified through segmentation. In addition, we perform an in-depth interpretability analysis of the model, proposing the Medical Explainability Index (MXI), a novel metric that quantifies the medical relevance of the generated heatmaps by comparing them with the model's lesion segmentation output. This metric provides a measurable basis to evaluate whether the model's decisions are grounded in clinically meaningful information. The proposed method was trained and evaluated on the Automatic Detection Challenge on Age-Related Macular Degeneration (ADAM) dataset. Experimental results demonstrate robust performance, achieving an area under the curve (AUC) of 0.96 for classification and a Dice similarity coefficient (DSC) of 0.59 for segmentation, outperforming single-task models. By offering interpretable and clinically relevant insights, our approach aims to foster greater trust in AI-driven disease diagnosis and facilitate its adoption in clinical practice.

We conducted a systematic review and meta-analysis to evaluate the performance of magnetic resonance imaging (MRI)-derived deep learning (DL) models in predicting 1p/19q codeletion status in glioma patients. The literature search was performed in four databases: PubMed, Web of Science, Embase, and Scopus. We included the studies that evaluated the performance of end-to-end DL models in predicting the status of glioma 1p/19q codeletion. The quality of the included studies was assessed by the Quality assessment of diagnostic accuracy studies-2 (QUADAS-2) METhodological RadiomICs Score (METRICS). We calculated diagnostic pooled estimates and heterogeneity was evaluated using I². Subgroup analysis and sensitivity analysis were conducted to explore sources of heterogeneity. Publication bias was evaluated by Deeks' funnel plots. Twenty studies were included in the systematic review. Only two studies had a low quality. A meta-analysis of the ten studies demonstrated a pooled sensitivity of 0.77 (95% CI: 0.63-0.87), a specificity of 0.85 (95% CI: 0.74-0.92), a positive diagnostic likelihood ratio (DLR) of 5.34 (95% CI: 2.88-9.89), a negative DLR of 0.26 (95% CI: 0.16-0.45), a diagnostic odds ratio of 20.24 (95% CI: 8.19-50.02), and an area under the curve of 0.89 (95% CI: 0.86-0.91). The subgroup analysis identified a significant difference between groups depending on the segmentation method used. DL models can predict glioma 1p/19q codeletion status with high accuracy and may enhance non-invasive tumor characterization and aid in the selection of optimal therapeutic strategies.

Information overload in electronic health records requires effective solutions to alleviate clinicians' administrative tasks. Automatically summarizing clinical text has gained significant attention with the rise of large language models. While individual studies show optimism, a structured overview of the research landscape is lacking. This study aims to present the current state of the art on clinical text summarization using large language models, evaluate the level of evidence in existing research and assess the applicability of performance findings in clinical settings. This scoping review complied with the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews) guidelines. Literature published between January 1, 2019, and June 18, 2024, was identified from 5 databases: PubMed, Embase, Web of Science, IEEE Xplore, and ACM Digital Library. Studies were excluded if they did not describe transformer-based models, did not focus on clinical text summarization, did not engage with free-text data, were not original research, were nonretrievable, were not peer-reviewed, or were not in English, French, Spanish, or German. Data related to study context and characteristics, scope of research, and evaluation methodologies were systematically collected and analyzed by 3 authors independently. A total of 30 original studies were included in the analysis. All used observational retrospective designs, mainly using real patient data (n=28, 93%). The research landscape demonstrated a narrow research focus, often centered on summarizing radiology reports (n=17, 57%), primarily involving data from the intensive care unit (n=15, 50%) of US-based institutions (n=19, 73%), in English (n=26, 87%). This focus aligned with the frequent reliance on the open-source Medical Information Mart for Intensive Care dataset (n=15, 50%). Summarization methodologies predominantly involved abstractive approaches (n=17, 57%) on single-document inputs (n=4, 13%) with unstructured data (n=13, 43%), yet reporting on methodological details remained inconsistent across studies. Model selection involved both open-source models (n=26, 87%) and proprietary models (n=7, 23%). Evaluation frameworks were highly heterogeneous. All studies conducted internal validation, but external validation (n=2, 7%), failure analysis (n=6, 20%), and patient safety risks analysis (n=1, 3%) were infrequent, and none reported bias assessment. Most studies used both automated metrics and human evaluation (n=16, 53%), while 10 (33%) used only automated metrics, and 4 (13%) only human evaluation. Key barriers hinder the translation of current research into trustworthy, clinically valid applications. Current research remains exploratory and limited in scope, with many applications yet to be explored. Performance assessments often lack reliability, and clinical impact evaluations are insufficient raising concerns about model utility, safety, fairness, and data privacy. Advancing the field requires more robust evaluation frameworks, a broader research scope, and a stronger focus on real-world applicability.

Introduction: Timely identification of intracranial hemorrhage (ICH) subtypes on non-contrast computed tomography is critical for prognosis prediction and therapeutic decision-making, yet remains challenging due to low contrast and blurring boundaries. This study evaluates the performance of zero-shot multi-modal large language models (MLLMs) compared to traditional deep learning methods in ICH binary classification and subtyping. Methods: We utilized a dataset provided by RSNA, comprising 192 NCCT volumes. The study compares various MLLMs, including GPT-4o, Gemini 2.0 Flash, and Claude 3.5 Sonnet V2, with conventional deep learning models, including ResNet50 and Vision Transformer. Carefully crafted prompts were used to guide MLLMs in tasks such as ICH presence, subtype classification, localization, and volume estimation. Results: The results indicate that in the ICH binary classification task, traditional deep learning models outperform MLLMs comprehensively. For subtype classification, MLLMs also exhibit inferior performance compared to traditional deep learning models, with Gemini 2.0 Flash achieving an macro-averaged precision of 0.41 and a macro-averaged F1 score of 0.31. Conclusion: While MLLMs excel in interactive capabilities, their overall accuracy in ICH subtyping is inferior to deep networks. However, MLLMs enhance interpretability through language interactions, indicating potential in medical imaging analysis. Future efforts will focus on model refinement and developing more precise MLLMs to improve performance in three-dimensional medical image processing.

Evaluate an Artificial Intelligence (AI) system in breast screening through stratified results across age, breast density, ethnicity and screening centres, from different UK regions. A large-scale retrospective study evaluating two variations of using AI as an independent second reader in double reading was executed. Stratifications were conducted for clinical and operational metrics. Data from 306 839 mammography cases screened between 2017 and 2021 were used and included three different UK regions.The impact on safety and effectiveness was assessed using clinical metrics: cancer detection rate and positive predictive value, stratified according to age, breast density and ethnicity. Operational impact was assessed through reading workload and recall rate, measured overall and per centre.Non-inferiority was tested for AI workflows compared with human double reading, and when passed, superiority was tested. AI interval cancer (IC) flag rate was assessed to estimate additional cancer detection opportunity with AI that cannot be assessed retrospectively. The AI workflows passed non-inferiority or superiority tests for every metric across all subgroups, with workload savings between 38.3% and 43.7%. The AI standalone flagged 41.2% of ICs overall, ranging between 33.3% and 46.8% across subgroups, with the highest detection rate for dense breasts. Human double reading and AI workflows showed the same performance disparities across subgroups. The AI integrations maintained or improved performance at all metrics for all subgroups while achieving significant workload reduction. Moreover, complementing these integrations with AI as an additional reader can improve cancer detection. The granularity of assessment showed that screening with the AI-system integrations was as safe as standard double reading across heterogeneous populations.

Positron emission tomography with 2-deoxy-2-[fluorine-18]fluoro-D-glucose integrated with computed tomography (18F-FDG PET-CT) is a multi-modality medical imaging technique widely used for screening and diagnosis of lesions and tumors, in which, CT can provide detailed anatomical structures, while PET can show metabolic activities. Nevertheless, it has disadvantages such as long scanning time, high cost, and relatively high radiation doses.&#xD;&#xD;Purpose: We propose a deep learning model for the whole-body CT-to-PET synthesis task, generating high-quality synthetic PET images that are comparable to real ones in both clinical relevance and diagnostic value.&#xD;&#xD;Material: We collect 102 pairs of 3D CT and PET scans, which are sliced into 27,240 pairs of 2D CT and PET images ( training: 21,855 pairs, validation: 2,810, testing: 2,575 pairs).&#xD;&#xD;Methods: We propose a Transformer-enhanced Generative Adversarial Network (GAN) for whole-body CT-to-PET synthesis task. The CPGAN model uses residual blocks and Fully Connected Transformer Residual (FCTR) blocks to capture both local features and global contextual information. A customized loss function incorporating structural consistency is designed to improve the quality of synthesized PET images.&#xD;&#xD;Results: Both quantitative and qualitative evaluation results demonstrate effectiveness of the CPGAN model. The mean and standard variance of NRMSE,PSNR and SSIM values on test set are (16.90 ± 12.27) × 10-4, 28.71 ± 2.67 and 0.926 ± 0.033, respectively, outperforming other seven state-of-the-art models. Three radiologists independently and blindly evaluated and gave subjective scores to 100 randomly chosen PET images (50 real and 50 synthetic). By Wilcoxon signed rank test, there are no statistical differences between the synthetic PET images and the real ones.&#xD;&#xD;Conclusions: Despite the inherent limitations of CT images to directly reflect biological information of metabolic tissues, CPGAN model effectively synthesizes satisfying PET images from CT scans, which has potential in reducing the reliance on actual PET-CT scans.

Out-of-distribution (OOD) detection is essential for ensuring the reliability of deep learning models in medical imaging applications. This work is motivated by the observation that class activation maps (CAMs) for in-distribution (ID) data typically emphasize regions that are highly relevant to the model's predictions, whereas OOD data often lacks such focused activations. By masking input images with inverted CAMs, the feature representations of ID data undergo more substantial changes compared to those of OOD data, offering a robust criterion for differentiation. In this paper, we introduce a novel unsupervised OOD detection framework, Multi-Exit Class Activation Map (MECAM), which leverages multi-exit CAMs and feature masking. By utilizing mult-exit networks that combine CAMs from varying resolutions and depths, our method captures both global and local feature representations, thereby enhancing the robustness of OOD detection. We evaluate MECAM on multiple ID datasets, including ISIC19 and PathMNIST, and test its performance against three medical OOD datasets, RSNA Pneumonia, COVID-19, and HeadCT, and one natural image OOD dataset, iSUN. Comprehensive comparisons with state-of-the-art OOD detection methods validate the effectiveness of our approach. Our findings emphasize the potential of multi-exit networks and feature masking for advancing unsupervised OOD detection in medical imaging, paving the way for more reliable and interpretable models in clinical practice.

Deep learning (DL) has become the dominant approach for medical image segmentation, yet ensuring the reliability and clinical applicability of these models requires addressing key challenges such as annotation variability, calibration, and uncertainty estimation. This is why we created the Calibration and Uncertainty for multiRater Volume Assessment in multiorgan Segmentation (CURVAS), which highlights the critical role of multiple annotators in establishing a more comprehensive ground truth, emphasizing that segmentation is inherently subjective and that leveraging inter-annotator variability is essential for robust model evaluation. Seven teams participated in the challenge, submitting a variety of DL models evaluated using metrics such as Dice Similarity Coefficient (DSC), Expected Calibration Error (ECE), and Continuous Ranked Probability Score (CRPS). By incorporating consensus and dissensus ground truth, we assess how DL models handle uncertainty and whether their confidence estimates align with true segmentation performance. Our findings reinforce the importance of well-calibrated models, as better calibration is strongly correlated with the quality of the results. Furthermore, we demonstrate that segmentation models trained on diverse datasets and enriched with pre-trained knowledge exhibit greater robustness, particularly in cases deviating from standard anatomical structures. Notably, the best-performing models achieved high DSC and well-calibrated uncertainty estimates. This work underscores the need for multi-annotator ground truth, thorough calibration assessments, and uncertainty-aware evaluations to develop trustworthy and clinically reliable DL-based medical image segmentation models.