Large language models (LLMs) are increasingly used by physicians for diagnostic support. A key advantage of LLMs is the ability to generate explanations that can help physicians understand the reasoning behind a diagnosis. However, the best-suited format for LLM-generated explanations remains unclear. In this large-scale study, we examined the effect of different formats for LLM explanations on clinical decision-making. For this, we conducted a randomized experiment with radiologists reviewing patient cases with radiological images (N = 2020 assessments). Participants received either no LLM support (control group) or were supported by one of three LLM-generated explanations: (1) a standard output providing the diagnosis without explanation; (2) a differential diagnosis comparing multiple possible diagnoses; or (3) a chain-of-thought explanation offering a detailed reasoning process for the diagnosis. We find that the format of explanations significantly influences diagnostic accuracy. The chain-of-thought explanations yielded the best performance, improving the diagnostic accuracy by 12.2% compared to the control condition without LLM support (P = 0.001). The chain-of-thought explanations are also superior to the standard output without explanation (+7.2%; P = 0.040) and the differential diagnosis format (+9.7%; P = 0.004). We further assessed the robustness of these findings across case difficulty and different physician backgrounds such as general vs. specialized radiologists. Evidently, explaining the reasoning for a diagnosis helps physicians to identify and correct potential errors in LLM predictions and thus improve overall decisions. Altogether, the results highlight the importance of how explanations in medical LLMs are generated to maximize their utility in clinical practice. By designing explanations to support the reasoning processes of physicians, LLMs can improve diagnostic performance and, ultimately, patient outcomes.

Breast cancer remains a significant global health issue. Despite advances in detection and treatment, its complexity is driven by genetic, environmental, and structural factors. Computational methods like Finite Element Modeling (FEM) have transformed our understanding of breast cancer risk and progression. Advanced computational approaches in breast cancer research are the focus, with an emphasis on FEM's role in simulating breast tissue mechanics and enhancing precision in therapies such as radiofrequency ablation (RFA). Machine learning (ML), particularly Convolutional Neural Networks (CNNs), has revolutionized imaging modalities like mammograms and MRIs, improving diagnostic accuracy and early detection. AI applications in analyzing histopathological images have advanced tumor classification and grading, offering consistency and reducing inter-observer variability. Explainability tools like Grad-CAM, SHAP, and LIME enhance the transparency of AI-driven models, facilitating their integration into clinical workflows. Integrating FEM and ML represents a paradigm shift in breast cancer management. FEM offers precise modeling of tissue mechanics, while ML excels in predictive analytics and image analysis. Despite challenges such as data variability and limited standardization, synergizing these approaches promises adaptive, personalized care. These computational methods have the potential to redefine diagnostics, optimize treatment, and improve patient outcomes.

To evaluate diagnostic accuracy and image quality of deep learning (DL) cine sequences for LV and RV parameters compared to conventional balanced steady-state free precession (bSSFP) cine sequences in cardiovascular magnetic resonance (CMR). From January to April 2024, patients with clinically indicated CMR were prospectively included. LV and RV were segmented from short-axis bSSFP and DL cine sequences. LV and RV end-diastolic volume (EDV), end-systolic volume (EDV), stroke volume (SV), ejection fraction, and LV end-diastolic mass were calculated. The acquisition time of both sequences was registered. Results were compared with paired-samples t test or Wilcoxon signed-rank test. Agreement between DL cine and bSSFP was assessed using Bland-Altman plots. Image quality was graded by two readers based on blood-to-myocardium contrast, endocardial edge definition, and motion artifacts, using a 5-point Likert scale (1 = insufficient quality; 5 = excellent quality). Sixty-two patients were included (mean age: 47 ± 17 years, 41 men). No significant differences between DL cine and bSSFP were found for all LV and RV parameters (P ≥ .176). DL cine was significantly faster (1.35 ± .55 m vs 2.83 ± .79 m; P < .001). The agreement between DL cine and bSSFP was strong, with bias ranging from 45 to 1.75% for LV and from - 0.38 to 2.43% for RV. Among LV parameters, the highest agreement was obtained for ESV and SV, which fell within the acceptable limit of agreement (LOA) in 84% of cases. EDV obtained the highest agreement among RV parameters, falling within the acceptable LOA in 90% of cases. Overall image quality was comparable (median: 5, IQR: 4-5; P = .330), while endocardial edge definition of DL cine (median: 4, IQR: 4-5) was lower than bSSFP (median: 5, IQR: 4-5; P = .002). DL cine allows fast and accurate quantification of LV and RV parameters and comparable image quality with conventional bSSFP.

This study aimed to develop deep learning algorithms (DLAs) utilising convolutional neural networks (CNNs) to classify cervical spondylotic myelopathy (CSM) and cervical spondylotic radiculopathy (CSR) from plain cervical spine radiographs. Data from 300 patients (150 with CSM and 150 with CSR) were used for internal validation (IV) using five-fold cross-validation strategy. Additionally, 100 patients (50 with CSM and 50 with CSR) were included in the external validation (EV). Two DLAs were trained using CNNs on plain radiographs from C3-C6 for the binary classification of CSM and CSR, and for the prediction of the spinal canal area rate using magnetic resonance imaging. Model performance was evaluated on external data using metrics such as area under the curve (AUC), accuracy, and likelihood ratios. For the binary classification, the AUC ranged from 0.84 to 0.96, with accuracy between 78% and 95% during IV. In the EV, the AUC and accuracy were 0.96 and 90%, respectively. For the spinal canal area rate, correlation coefficients during five-fold cross-validation ranged from 0.57 to 0.64, with a mean correlation of 0.61 observed in the EV. DLAs developed with CNNs demonstrated promising accuracy for classifying CSM and CSR from plain radiographs. These algorithms have the potential to assist non-specialists in identifying patients who require further evaluation or referral to spine specialists, thereby reducing delays in the diagnosis and treatment of CSM.

Artifacts in clinical MRI can compromise the performance of AI models. This study evaluates how different data augmentation strategies affect an AI model's segmentation performance under variable artifact severity. We used an AI model based on the nnU-Net architecture to automatically quantify lower limb alignment using axial T2-weighted MR images. Three versions of the AI model were trained with different augmentation strategies: (1) no augmentation ("baseline"), (2) standard nnU-net augmentations ("default"), and (3) "default" plus augmentations that emulate MR artifacts ("MRI-specific"). Model performance was tested on 600 MR image stacks (right and left; hip, knee, and ankle) from 20 healthy participants (mean age, 23 ± 3 years, 17 men), each imaged five times under standardized motion to induce artifacts. Two radiologists graded each stack's artifact severity as none, mild, moderate, and severe, and manually measured torsional angles. Segmentation quality was assessed using the Dice similarity coefficient (DSC), while torsional angles were compared between manual and automatic measurements using mean absolute deviation (MAD), intraclass correlation coefficient (ICC), and Pearson's correlation coefficient (r). Statistical analysis included parametric tests and a Linear Mixed-Effects Model. MRI-specific augmentation resulted in slightly (yet not significantly) better performance than the default strategy. Segmentation quality decreased with increasing artifact severity, which was partially mitigated by default and MRI-specific augmentations (e.g., severe artifacts, proximal femur: DSC_baseline = 0.58 ± 0.22; DSC_default = 0.72 ± 0.22; DSC_MRI-specific = 0.79 ± 0.14 [p < 0.001]). These augmentations also maintained precise torsional angle measurements (e.g., severe artifacts, femoral torsion: MAD_baseline = 20.6 ± 23.5°; MAD_default = 7.0 ± 13.0°; MAD_MRI-specific = 5.7 ± 9.5° [p < 0.001]; ICC_baseline = -0.10 [p = 0.63; 95% CI: -0.61 to 0.47]; ICC_default = 0.38 [p = 0.08; -0.17 to 0.76]; ICC_MRI-specific = 0.86 [p < 0.001; 0.62 to 0.95]; r_baseline = 0.58 [p < 0.001; 0.44 to 0.69]; r_default = 0.68 [p < 0.001; 0.56 to 0.77]; r_MRI-specific = 0.86 [p < 0.001; 0.81 to 0.9]). Motion artifacts negatively impact AI models, but general-purpose augmentations enhance robustness effectively. MRI-specific augmentations offer minimal additional benefit. Question Motion artifacts negatively impact the performance of diagnostic AI models for MRI, but mitigation methods remain largely unexplored. Findings Domain-specific augmentation during training can improve the robustness and performance of a model for quantifying lower limb alignment in the presence of severe artifacts. Clinical relevance Excellent robustness and accuracy are crucial for deploying diagnostic AI models in clinical practice. Including domain knowledge in model training can benefit clinical adoption.

Voice disorders (VD) are often linked to vocal fold structural pathologies (VFSP). Laryngeal imaging plays a vital role in assessing VFSPs and VD in clinical and research settings, but challenges like scarce and imbalanced datasets can limit the generalizability of findings. Denoising Diffusion Probabilistic Models (DDPMs), a subtype of Generative AI, has gained attention for its ability to generate high-quality and realistic synthetic images to address these challenges. This study explores the feasibility of improving VFSP image classification by generating synthetic images using DDPMs. 404 laryngoscopic images depicting VF without and with VFSP were included. DDPMs were used to generate synthetic images to augment the original dataset. Two convolutional neural network architectures, VGG16 and ResNet50, were applied for model training. The models were initially trained only on the original dataset. Then, they were trained on the augmented datasets. Evaluation metrics were analyzed to assess the performance of the models for both binary classification (with/without VFSPs) and multi-class classification (seven specific VFSPs). Realistic and high-quality synthetic images were generated for dataset augmentation. The model first failed to converge when trained only on the original dataset, but they successfully converged and achieved low loss and high accuracy when trained on the augmented datasets. The best performance was gained for both binary and multi-class classification when the models were trained on an augmented dataset. Generating realistic images of VFSP using DDPMs is feasible and can enhance the classification of VFSPs by an AI model and may support VD screening and diagnosis.

Deep learning techniques have significantly enhanced the convenience and precision of ultrasound image diagnosis, particularly in the crucial step of lesion segmentation. However, recent studies reveal that both train-from-scratch models and pre-trained models often exhibit performance disparities across sex and age attributes, leading to biased diagnoses for different subgroups. In this paper, we propose APPLE, a novel approach designed to mitigate unfairness without altering the parameters of the base model. APPLE achieves this by learning fair perturbations in the latent space through a generative adversarial network. Extensive experiments on both a publicly available dataset and an in-house ultrasound image dataset demonstrate that our method improves segmentation and diagnostic fairness across all sensitive attributes and various backbone architectures compared to the base models. Through this study, we aim to highlight the critical importance of fairness in medical segmentation and contribute to the development of a more equitable healthcare system.

High-grade clear cell renal cell carcinoma (ccRCC) is linked to lower survival rates and more aggressive disease progression. This study aims to assess the diagnostic performance of MRI-derived radiomics as a non-invasive approach for pre-operative differentiation of high-grade from low-grade ccRCC. A systematic search was conducted across PubMed, Scopus, and Embase. Quality assessment was performed using QUADAS-2 and METRICS. Pooled sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under the curve (AUC) were estimated using a bivariate model. Separate meta-analyses were conducted for radiomics models and combined models, where the latter integrated clinical and radiological features with radiomics. Subgroup analysis was performed to identify potential sources of heterogeneity. Sensitivity analysis was conducted to identify potential outliers. A total of 15 studies comprising 2,265 patients were included, with seven and six studies contributing to the meta-analysis of radiomics and combined models, respectively. The pooled estimates of the radiomics model were as follows: sensitivity, 0.78; specificity, 0.84; PLR, 4.17; NLR, 0.28; DOR, 17.34; and AUC, 0.84. For the combined model, the pooled sensitivity, specificity, PLR, NLR, DOR, and AUC were 0.87, 0.81, 3.78, 0.21, 28.57, and 0.90, respectively. Radiomics models trained on smaller cohorts exhibited a significantly higher pooled specificity and PLR than those trained on larger cohorts. Also, radiomics models based on single-user segmentation demonstrated a significantly higher pooled specificity compared to multi-user segmentation. Radiomics has demonstrated potential as a non-invasive tool for grading ccRCC, with combined models achieving superior performance.

This study aims to develop a deep learning algorithm for differentiating aneurysmal subarachnoid hemorrhage (aSAH) from non-aneurysmal subarachnoid hemorrhage (naSAH) using non-contrast computed tomography (NCCT) scans. This retrospective study included 618 patients diagnosed with SAH. The dataset was divided into a training and internal validation cohort (533 cases: aSAH = 305, naSAH = 228) and an external test cohort (85 cases: aSAH = 55, naSAH = 30). Hemorrhage regions were automatically segmented using a U-Net + + architecture. A ResNet-based deep learning model was trained to classify the etiology of SAH. The model achieved robust performance in distinguishing aSAH from naSAH. In the internal validation cohort, it yielded an average sensitivity of 0.898, specificity of 0.877, accuracy of 0.889, Matthews correlation coefficient (MCC) of 0.777, and an area under the curve (AUC) of 0.948 (95% CI: 0.929-0.967). In the external test cohort, the model demonstrated an average sensitivity of 0.891, specificity of 0.880, accuracy of 0.887, MCC of 0.761, and AUC of 0.914 (95% CI: 0.889-0.940), outperforming junior radiologists (average accuracy: 0.836; MCC: 0.660). The study presents a deep learning architecture capable of accurately identifying SAH etiology from NCCT scans. The model's high diagnostic performance highlights its potential to support rapid and precise clinical decision-making in emergency settings. Question Differentiating aneurysmal from naSAH is crucial for timely treatment, yet existing imaging modalities are not universally accessible or convenient for rapid diagnosis. Findings A ResNet-variant-based deep learning model utilizing non-contrast CT scans demonstrated high accuracy in classifying SAH etiology and enhanced junior radiologists' diagnostic performance. Clinical relevance AI-driven analysis of non-contrast CT scans provides a fast, cost-effective, and non-invasive solution for preoperative SAH diagnosis. This approach facilitates early identification of patients needing aneurysm surgery while minimizing unnecessary angiography in non-aneurysmal cases, enhancing clinical workflow efficiency.

By helping the neurosurgeon create treatment strategies that increase the survival rate, automotive diagnosis and CT (Computed Tomography) hemorrhage segmentation (CT) could be beneficial. Owing to the significance of medical image segmentation and the difficulties in carrying out human operations, a wide variety of automated techniques for this purpose have been developed, with a primary focus on particular image modalities. In this paper, MUNet (Multiclass-UNet) based Intracranial Hemorrhage Segmentation and Classification Framework (IHSNet) is proposed to successfully segment multiple kinds of hemorrhages while the fully connected layers help in classifying the type of hemorrhages.The segmentation accuracy rates for hemorrhages are 98.53% with classification accuracy stands at 98.71% when using the suggested approach. There is potential for this suggested approach to be expanded in the future to handle further medical picture segmentation issues. Intraventricular hemorrhage (IVH), Epidural hemorrhage (EDH), Intraparenchymal hemorrhage (IPH), Subdural hemorrhage (SDH), Subarachnoid hemorrhage (SAH) are the subtypes involved in intracranial hemorrhage (ICH) whose DICE coefficients are 0.77, 0.84, 0.64, 0.80, and 0.92 respectively.The proposed method has great deal of clinical application potential for computer-aided diagnostics, which can be expanded in the future to handle further medical picture segmentation and to tackle with the involved issues.