Trauma-induced rib fractures are common injuries. The gold standard for diagnosing rib fractures is computed tomography (CT), but the sensitivity in the acute setting is low, and interpreting CT slices is labor-intensive. This has led to the development of new diagnostic approaches leveraging deep learning (DL) models. This systematic review and pooled analysis aimed to compare the performance of DL models in the detection, segmentation, and classification of rib fractures based on CT scans. A literature search was performed using various databases for studies describing DL models detecting, segmenting, or classifying rib fractures from CT data. Reported performance metrics included sensitivity, false-positive rate, F1-score, precision, accuracy, and mean average precision. A meta-analysis was performed on the sensitivity scores to compare the DL models with clinicians. Of the 323 identified records, 25 were included. Twenty-one studies reported on detection, four on segmentation, and 10 on classification. Twenty studies had adequate data for meta-analysis. The gold standard labels were provided by clinicians who were radiologists and orthopedic surgeons. For detecting rib fractures, DL models had a higher sensitivity (86.7%; 95% CI: 82.6%-90.2%) than clinicians (75.4%; 95% CI: 68.1%-82.1%). In classification, the sensitivity of DL models for displaced rib fractures (97.3%; 95% CI: 95.6%-98.5%) was significantly better than that of clinicians (88.2%; 95% CI: 84.8%-91.3%). DL models for rib fracture detection and classification achieved promising results. With better sensitivities than clinicians for detecting and classifying displaced rib fractures, the future should focus on implementing DL models in daily clinics. Level III-systematic review and pooled analysis.

Stroke is among the top three causes of death worldwide, and accurate identification of stroke lesion boundaries is critical for diagnosis and treatment. Supervised deep learning methods have emerged as the leading solution for stroke lesion segmentation but require large, diverse, and annotated datasets. The ISLES'24 challenge addresses this need by providing longitudinal stroke imaging data, including CT scans taken on arrival to the hospital and follow-up MRI taken 2-9 days from initial arrival, with annotations derived from follow-up MRI. Importantly, models submitted to the ISLES'24 challenge are evaluated using only CT inputs, requiring prediction of lesion progression that may not be visible in CT scans for segmentation. Our winning solution shows that a carefully designed preprocessing pipeline including deep-learning-based skull stripping and custom intensity windowing is beneficial for accurate segmentation. Combined with a standard large residual nnU-Net architecture for segmentation, this approach achieves a mean test Dice of 28.5 with a standard deviation of 21.27.

Background: Accurate MRI-based identification of extramural vascular invasion (EVI) and mesorectal fascia invasion (MFI) is pivotal for risk-stratified management of rectal cancer, yet visual assessment is subjective and vulnerable to inter-institutional variability. Purpose: To develop and externally evaluate a multicenter, foundation-model-driven framework that automatically classifies EVI and MFI on axial and sagittal T2-weighted MRI. Methods: This retrospective study used 331 pre-treatment rectal cancer MRI examinations from three European hospitals. After TotalSegmentator-guided rectal patch extraction, a self-supervised frequency-domain harmonization pipeline was trained to minimize scanner-related contrast shifts. Four classifiers were compared: ResNet50, SeResNet, the universal biomedical pretrained transformer (UMedPT) with a lightweight MLP head, and a logistic-regression variant using frozen UMedPT features (UMedPT_LR). Results: UMedPT_LR achieved the best EVI detection when axial and sagittal features were fused (AUC = 0.82; sensitivity = 0.75; F1 score = 0.73), surpassing the Chaimeleon Grand-Challenge winner (AUC = 0.74). The highest MFI performance was attained by UMedPT on axial harmonized images (AUC = 0.77), surpassing the Chaimeleon Grand-Challenge winner (AUC = 0.75). Frequency-domain harmonization improved MFI classification but variably affected EVI performance. Conventional CNNs (ResNet50, SeResNet) underperformed, especially in F1 score and balanced accuracy. Conclusion: These findings demonstrate that combining foundation model features, harmonization, and multi-view fusion significantly enhances diagnostic performance in rectal MRI.

Deep learning has transformed computer vision but relies heavily on large labeled datasets and computational resources. Transfer learning, particularly fine-tuning pretrained models, offers a practical alternative; however, models pretrained on natural image datasets such as ImageNet may fail to capture domain-specific characteristics in medical imaging. This study introduces an unsupervised learning framework that extracts high-value dermatological features instead of relying solely on ImageNet-based pretraining. We employ a Variational Autoencoder (VAE) trained from scratch on a proprietary dermatological dataset, allowing the model to learn a structured and clinically relevant latent space. This self-supervised feature extractor is then compared to an ImageNet-pretrained backbone under identical classification conditions, highlighting the trade-offs between general-purpose and domain-specific pretraining. Our results reveal distinct learning patterns. The self-supervised model achieves a final validation loss of 0.110 (-33.33%), while the ImageNet-pretrained model stagnates at 0.100 (-16.67%), indicating overfitting. Accuracy trends confirm this: the self-supervised model improves from 45% to 65% (+44.44%) with a near-zero overfitting gap, whereas the ImageNet-pretrained model reaches 87% (+50.00%) but plateaus at 75% (+19.05%), with its overfitting gap increasing to +0.060. These findings suggest that while ImageNet pretraining accelerates convergence, it also amplifies overfitting on non-clinically relevant features. In contrast, self-supervised learning achieves steady improvements, stronger generalization, and superior adaptability, underscoring the importance of domain-specific feature extraction in medical imaging.

The aim of this study is to develop a deep learning-based multimodal feature interaction-guided fusion (DL-MFIF) framework that integrates macroscopic information from computed tomography (CT) images with microscopic information from whole-slide images (WSIs) to predict the epidermal growth factor receptor (EGFR) mutations of primary lung adenocarcinoma in patients with advanced-stage disease. Data from 396 patients with lung adenocarcinoma across two medical institutions were analyzed. The data from 243 cases were divided into a training set (n=145) and an internal validation set (n=98) in a 6:4 ratio, and data from an additional 153 cases from another medical institution were included as an external validation set. All cases included CT scan images and WSIs. To integrate multimodal information, we developed the DL-MFIF framework, which leverages deep learning techniques to capture the interactions between radiomic macrofeatures derived from CT images and microfeatures obtained from WSIs. Compared to other classification models, the DL-MFIF model achieved significantly higher area under the curve (AUC) values. Specifically, the model outperformed others on both the internal validation set (AUC=0.856, accuracy=0.750) and the external validation set (AUC=0.817, accuracy=0.708). Decision curve analysis (DCA) demonstrated that the model provided superior net benefits(range 0.15-0.87). Delong's test for external validation confirmed the statistical significance of the results (P<0.05). The DL-MFIF model demonstrated excellent performance in evaluating and distinguishing the EGFR in patients with advanced lung adenocarcinoma. This model effectively aids radiologists in accurately classifying EGFR mutations in patients with primary lung adenocarcinoma, thereby improving treatment outcomes for this population.

To evaluate the variability of fully automated airway quantitative CT (QCT) measures caused by different kernels and the effect of kernel conversion. This retrospective study included 96 patients who underwent non-enhanced chest CT at two centers. CT scans were reconstructed using four kernels (medium soft, medium sharp, sharp, very sharp) from three vendors. Kernel conversion targeting the medium soft kernel as reference was applied to sharp kernel images. Fully automated airway quantification was performed before and after conversion. The effects of kernel type and conversion on airway quantification were evaluated using analysis of variance, paired t-tests, and concordance correlation coefficient (CCC). Airway QCT measures (e.g., Pi10, wall thickness, wall area percentage, lumen diameter) decreased with sharper kernels (all, p < 0.001), with varying degrees of variability across variables and vendors. Kernel conversion substantially reduced variability between medium soft and sharp kernel images for vendors A (pooled CCC: 0.59 vs. 0.92) and B (0.40 vs. 0.91) and lung-dedicated sharp kernels of vendor C (0.26 vs. 0.71). However, it was ineffective for non-lung-dedicated sharp kernels of vendor C (0.81 vs. 0.43) and showed limited improvement in variability of QCT measures at the subsegmental level. Consistent airway segmentation and identical anatomic labeling improved subsegmental airway variability in theoretical tests. Deep learning-based kernel conversion reduced the measurement variability of airway QCT across various kernels and vendors but was less effective for non-lung-dedicated kernels and subsegmental airways. Consistent airway segmentation and precise anatomic labeling can further enhance reproducibility for reliable automated quantification. Question How do different CT reconstruction kernels affect the measurement variability of automated airway measurements, and can deep learning-based kernel conversion reduce this variability? Findings Kernel conversion improved measurement consistency across vendors for lung-dedicated kernels, but showed limited effectiveness for non-lung-dedicated kernels and subsegmental airways. Clinical relevance Understanding kernel-related variability in airway quantification and mitigating it through deep learning enables standardized analysis, but further refinements are needed for robust airway segmentation, particularly for improving measurement variability in subsegmental airways and specific kernels.

As machine learning (ML) continuously develops in cancer diagnosis and treatment, some researchers have attempted to predict the expression of programmed death ligand-1 (PD-L1) in non-small cell lung cancer (NSCLC) by ML. However, there is a lack of systematic evidence on the effectiveness of ML. We conducted a thorough search across Embase, PubMed, the Cochrane Library, and Web of Science from inception to December 14th, 2023.A systematic review and meta-analysis was conducted to assess the value of ML for predicting PD-L1 expression in NSCLC. Totally 30 studies with 12,898 NSCLC patients were included. The thresholds of PD-L1 expression level were < 1%, 1-49%, and ≥ 50%. In the validation set, in the binary classification for PD-L1 ≥ 1%, the pooled C-index was 0.646 (95%CI: 0.587-0.705), 0.799 (95%CI: 0.782-0.817), 0.806 (95%CI: 0.753-0.858), and 0.800 (95%CI: 0.717-0.883), respectively, for the clinical feature-, radiomics-, radiomics + clinical feature-, and pathomics-based ML models; in the binary classification for PD-L1 ≥ 50%, the pooled C-index was 0.649 (95%CI: 0.553-0.744), 0.771 (95%CI: 0.728-0.814), and 0.826 (95%CI: 0.783-0.869), respectively, for the clinical feature-, radiomics-, and radiomics + clinical feature-based ML models. At present, radiomics- or pathomics-based ML methods are applied for the prediction of PD-L1 expression in NSCLC, which both achieve satisfactory accuracy. In particular, the radiomics-based ML method seems to have wider clinical applicability as a non-invasive diagnostic tool. Both radiomics and pathomics serve as processing methods for medical images. In the future, we expect to develop medical image-based DL methods for intelligently predicting PD-L1 expression.

<i>Objective.</i>This study investigates key factors influencing deep learning-based dose prediction models for head and neck cancer radiation therapy. The goal is to evaluate model accuracy, robustness, and computational efficiency, and to identify key components necessary for optimal performance.<i>Approach.</i>We systematically analyze the impact of input and dose grid resolution, input type, loss function, model architecture, and noise on model performance. Two datasets are used: a public dataset (OpenKBP) and an in-house clinical dataset. Model performance is primarily evaluated using two metrics: dose score and dose-volume histogram (DVH) score.<i>Main results.</i>High-resolution inputs improve prediction accuracy (dose score and DVH score) by 8.6%-13.5% compared to low resolution. Using a combination of CT, planning target volumes, and organs-at-risk as input significantly enhances accuracy, with improvements of 57.4%-86.8% over using CT alone. Integrating mean absolute error (MAE) loss with value-based and criteria-based DVH loss functions further boosts DVH score by 7.2%-7.5% compared to MAE loss alone. In the robustness analysis, most models show minimal degradation under Poisson noise (0-0.3 Gy) but are more susceptible to adversarial noise (0.2-7.8 Gy). Notably, certain models, such as SwinUNETR, demonstrate superior robustness against adversarial perturbations.<i>Significance.</i>These findings highlight the importance of optimizing deep learning models and provide valuable guidance for achieving more accurate and reliable radiotherapy dose prediction.

Heterogeneity is a fundamental characteristic of brain diseases, distinguished by variability not only in brain atrophy but also in the complexity of neural connectivity and brain networks. However, existing data-driven methods fail to provide a comprehensive analysis of brain heterogeneity. Recently, dynamic neural networks (DNNs) have shown significant advantages in capturing sample-wise heterogeneity. Therefore, in this article, we first propose a novel dynamic heterogeneity-aware network (DHANet) to identify critical heterogeneous brain regions, explore heterogeneous connectivity between them, and construct a heterogeneous-aware structural brain network (HGA-SBN) using structural magnetic resonance imaging (sMRI). Specifically, we develop a 3-D dynamic convmixer to extract abundant heterogeneous features from sMRI first. Subsequently, the critical brain atrophy regions are identified by dynamic prototype learning with embedding the hierarchical brain semantic structure. Finally, we employ a joint dynamic edge-correlation (JDE) modeling approach to construct the heterogeneous connectivity between these regions and analyze the HGA-SBN. To evaluate the effectiveness of the DHANet, we conduct elaborate experiments on three public datasets and the method achieves state-of-the-art (SOTA) performance on two classification tasks.

Prostate cancer (PCa) is the most commonly diagnosed malignancy in men worldwide. Accurate triage of patients based on tumor aggressiveness and staging is critical for selecting appropriate management pathways. While magnetic resonance imaging (MRI) has become a mainstay in PCa diagnosis, most predictive models rely on multiparametric imaging or invasive inputs, limiting generalizability in real-world clinical settings. This study aimed to develop and validate machine learning (ML) models using radiomic features extracted from T2-weighted MRI--alone and in combination with clinical variables--to predict ISUP grade (tumor aggressiveness), lymph node involvement (cN) and distant metastasis (cM). A retrospective multicenter cohort from three European sites in the Chaimeleon project was analyzed. Radiomic features were extracted from prostate zone segmentations and lesion masks, following standardized preprocessing and ComBat harmonization. Feature selection and model optimization were performed using nested cross-validation and Bayesian tuning. Hybrid models were trained using XGBoost and interpreted with SHAP values. The ISUP model achieved an AUC of 0.66, while the cN and cM models reached AUCs of 0.77 and 0.80, respectively. The best-performing models consistently combined prostate zone radiomics with clinical features such as PSA, PIRADSv2 and ISUP grade. SHAP analysis confirmed the importance of both clinical and texture-based radiomic features, with entropy and non-uniformity measures playing central roles in all tasks. Our results demonstrate the feasibility of using T2-weighted MRI and zonal radiomics for robust prediction of aggressiveness, nodal involvement and distant metastasis in PCa. This fully automated pipeline offers an interpretable, accessible and clinically translatable tool for first-line PCa triage, with potential integration into real-world diagnostic workflows.