Deep learning has driven significant advances in medical image analysis, yet its adoption in clinical practice remains constrained by the large size and lack of transparency in modern models. Advances in interpretability techniques such as DL-Backtrace, Layer-wise Relevance Propagation, and Integrated Gradients make it possible to assess the contribution of individual components within neural networks trained on medical imaging tasks. In this work, we introduce an interpretability-guided pruning framework that reduces model complexity while preserving both predictive performance and transparency. By selectively retaining only the most relevant parts of each layer, our method enables targeted compression that maintains clinically meaningful representations. Experiments across multiple medical image classification benchmarks demonstrate that this approach achieves high compression rates with minimal loss in accuracy, paving the way for lightweight, interpretable models suited for real-world deployment in healthcare settings.

Magnetic resonance imaging (MRI) enables non-invasive, high-resolution analysis of muscle structures. However, automated segmentation remains limited by high computational costs, reliance on large training datasets, and reduced accuracy in segmenting smaller muscles. Convolutional neural network (CNN)-based methods, while powerful, often suffer from substantial computational overhead, limited generalizability, and poor interpretability across diverse populations. This study proposes a training-free segmentation approach based on keypoint tracking, which integrates keypoint selection with Lucas-Kanade optical flow. The proposed method achieves a mean Dice similarity coefficient (DSC) ranging from 0.6 to 0.7, depending on the keypoint selection strategy, performing comparably to state-of-the-art CNN-based models while substantially reducing computational demands and enhancing interpretability. This scalable framework presents a robust and explainable alternative for muscle segmentation in clinical and research applications.

Recent works have revisited the infamous task ``Name That Dataset'', demonstrating that non-medical datasets contain underlying biases and that the dataset origin task can be solved with high accuracy. In this work, we revisit the same task applied to popular open-source chest X-ray datasets. Medical images are naturally more difficult to release for open-source due to their sensitive nature, which has led to certain open-source datasets being extremely popular for research purposes. By performing the same task, we wish to explore whether dataset bias also exists in these datasets. To extend our work, we apply simple transformations to the datasets, repeat the same task, and perform an analysis to identify and explain any detected biases. Given the importance of AI applications in medical imaging, it's vital to establish whether modern methods are taking shortcuts or are focused on the relevant pathology. We implement a range of different network architectures on the datasets: NIH, CheXpert, MIMIC-CXR and PadChest. We hope this work will encourage more explainable research being performed in medical imaging and the creation of more open-source datasets in the medical domain. Our code can be found here: https://github.com/eedack01/x_ray_ds_bias.

Recent work has revisited the infamous task Name that dataset and established that in non-medical datasets, there is an underlying bias and achieved high Accuracies on the dataset origin task. In this work, we revisit the same task applied to popular open-source chest X-ray datasets. Medical images are naturally more difficult to release for open-source due to their sensitive nature, which has led to certain open-source datasets being extremely popular for research purposes. By performing the same task, we wish to explore whether dataset bias also exists in these datasets. % We deliberately try to increase the difficulty of the task by dataset transformations. We apply simple transformations of the datasets to try to identify bias. Given the importance of AI applications in medical imaging, it's vital to establish whether modern methods are taking shortcuts or are focused on the relevant pathology. We implement a range of different network architectures on the datasets: NIH, CheXpert, MIMIC-CXR and PadChest. We hope this work will encourage more explainable research being performed in medical imaging and the creation of more open-source datasets in the medical domain. The corresponding code will be released upon acceptance.

Recent works have revisited the infamous task ``Name That Dataset'', demonstrating that non-medical datasets contain underlying biases and that the dataset origin task can be solved with high accuracy. In this work, we revisit the same task applied to popular open-source chest X-ray datasets. Medical images are naturally more difficult to release for open-source due to their sensitive nature, which has led to certain open-source datasets being extremely popular for research purposes. By performing the same task, we wish to explore whether dataset bias also exists in these datasets. To extend our work, we apply simple transformations to the datasets, repeat the same task, and perform an analysis to identify and explain any detected biases. Given the importance of AI applications in medical imaging, it's vital to establish whether modern methods are taking shortcuts or are focused on the relevant pathology. We implement a range of different network architectures on the datasets: NIH, CheXpert, MIMIC-CXR and PadChest. We hope this work will encourage more explainable research being performed in medical imaging and the creation of more open-source datasets in the medical domain. Our code can be found here: https://github.com/eedack01/x_ray_ds_bias.

Localized drug delivery can enhance therapeutic efficacy while minimizing systemic side effects, making sustained-release ophthalmic inserts an attractive alternative to traditional eye drops. Such inserts offer improved patient compliance through prolonged therapeutic effects and a reduced need for frequent administration. This study focuses on dexamethasone-containing ophthalmic inserts. These inserts utilize a key excipient, polyethylene glycol (PEG), which forms a hydrogel upon contact with tear fluid. Developing generic equivalents of PEG-based inserts is challenging due to difficulties in characterizing inactive ingredients and the absence of standardized physicochemical characterization methods to demonstrate similarity. To address this gap, a suite of analytical approaches was applied to both PEG precursor materials sourced from different vendors and manufactured inserts. ¹H NMR, FTIR, MALDI, and SEC revealed variations in end-group functionalization, impurity content, and molecular weight distribution of the excipient. These differences led to changes in the finished insert network properties such as porosity, pore size and structure, gel mechanical strength, and crystallinity, which were corroborated by X-ray microscopy, AI-based image analysis, thermal, mechanical, and density measurements. In vitro release testing revealed distinct drug release profiles across formulations, with swelling rate correlated to release rate (i.e., faster release with rapid swelling). The use of non-micronized and micronized dexamethasone also contributed to release profile differences. Through comprehensive characterization of these PEG-based dexamethasone inserts, correlations between polymer quality, hydrogel microstructure, and release kinetics were established. The study highlights how excipient differences can alter product performance, emphasizing the importance of thorough analysis in developing generic equivalents of complex drug products.

To establish an interpretable and non-invasive machine learning (ML) model using clinicoradiological predictors and magnetic resonance imaging (MRI) radiomics features to predict the consistency of pituitary macroadenomas (PMAs) preoperatively. Total 350 patients with PMA (272 from Xinqiao Hospital of Army Medical University and 78 from Daping Hospital of Army Medical University) were stratified and randomly divided into training and test cohorts in a 7:3 ratio. The tumor consistency was classified as soft or firm. Clinicoradiological predictors were examined utilizing univariate and multivariate regression analyses. Radiomics features were selected employing the minimum redundancy maximum relevance (mRMR) and least absolute shrinkage and selection operator (LASSO) algorithms. Logistic regression (LR) and random forest (RF) classifiers were applied to construct the models. Receiver operating characteristic (ROC) curves and decision curve analyses (DCA) were performed to compare and validate the predictive capacities of the models. A comparative study of the area under the curve (AUC), accuracy (ACC), sensitivity (SEN), and specificity (SPE) was performed. The Shapley additive explanation (SHAP) was applied to investigate the optimal model's interpretability. The combined model predicted the PMAs' consistency more effectively than the clinicoradiological and radiomics models. Specifically, the LR-combined model displayed optimal prediction performance (test cohort: AUC = 0.913; ACC = 0.840). The SHAP-based explanation of the LR-combined model suggests that the wavelet-transformed and Laplacian of Gaussian (LoG) filter features extracted from T₂WI and CE-T₁WI occupy a dominant position. Meanwhile, the skewness of the original first-order features extracted from T₂WI (T₂WI_original_first-order_Skewness) demonstrated the most substantial contribution. An interpretable machine learning model incorporating clinicoradiological predictors and multiparametric MRI (mpMRI)-based radiomics features may predict PMAs consistency, enabling tailored and precise therapies for patients with PMA.

The deployment of AI models in clinical practice faces a critical challenge: models achieving expert-level performance on benchmarks can fail catastrophically when confronted with real-world variations in medical imaging. Minor shifts in scanner hardware, lighting or demographics can erode accuracy, but currently reliability auditing to identify such catastrophic failure cases before deployment is a bespoke and time-consuming process. Practitioners lack accessible and interpretable tools to expose and repair hidden failure modes. Here we introduce ModelAuditor, a self-reflective agent that converses with users, selects task-specific metrics, and simulates context-dependent, clinically relevant distribution shifts. ModelAuditor then generates interpretable reports explaining how much performance likely degrades during deployment, discussing specific likely failure modes and identifying root causes and mitigation strategies. Our comprehensive evaluation across three real-world clinical scenarios - inter-institutional variation in histopathology, demographic shifts in dermatology, and equipment heterogeneity in chest radiography - demonstrates that ModelAuditor is able correctly identify context-specific failure modes of state-of-the-art models such as the established SIIM-ISIC melanoma classifier. Its targeted recommendations recover 15-25% of performance lost under real-world distribution shift, substantially outperforming both baseline models and state-of-the-art augmentation methods. These improvements are achieved through a multi-agent architecture and execute on consumer hardware in under 10 minutes, costing less than US$0.50 per audit.

Reducing radiation dose from PET imaging is essential to minimize cancer risks; however, it often leads to increased noise and degraded image quality, compromising diagnostic reliability. Recent advances in deep learning have shown promising results in addressing these limitations through effective denoising. However, existing networks trained on specific noise levels often fail to generalize across diverse acquisition conditions. Moreover, training multiple models for different noise levels is impractical due to data and computational constraints. This study aimed to develop a supervised Swin Transformer-based unified noise-aware (ST-UNN) network that handles diverse noise levels and reconstructs high-quality images in low-dose PET imaging. We present a Swin Transformer-based Noise-Aware Network (ST-UNN), which incorporates multiple sub-networks, each designed to address specific noise levels ranging from 1 % to 10 %. An adaptive weighting mechanism dynamically integrates the outputs of these sub-networks to achieve effective denoising. The model was trained and evaluated using PET/CT dataset encompassing the entire head and malignant lesions in the head and neck region. Performance was assessed using a combination of structural and statistical metrics, including the Structural Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR), Standardized Uptake Value (SUV) mean bias, SUV_max bias, and Root Mean Square Error (RMSE). This comprehensive evaluation ensured reliable results for both global and localized regions within PET images. The ST-UNN consistently outperformed conventional networks, particularly in ultra-low-dose scenarios. At 1 % count level, it achieved a PSNR of 34.77, RMSE of 0.05, and SSIM of 0.97, notably surpassing the baseline networks. It also achieved the lowest SUV_mean bias (0.08) and RMSE lesion (0.12) at this level. Across all count levels, ST-UNN maintained high performance and low error, demonstrating strong generalization and diagnostic integrity. ST-UNN offers a scalable, transformer-based solution for low-dose PET imaging. By dynamically integrating sub-networks, it effectively addresses noise variability and provides superior image quality, thereby advancing the capabilities of low-dose and dynamic PET imaging.

Ultrasound microvascular imaging (UMI) is often hindered by low signal-to-noise ratio (SNR), especially in contrast-free or deep tissue scenarios, which impairs subsequent vascular quantification and reliable disease diagnosis. To address this challenge, we propose Half-Angle-to-Half-Angle (HA2HA), a self-supervised denoising framework specifically designed for UMI. HA2HA constructs training pairs from complementary angular subsets of beamformed radio-frequency (RF) blood flow data, across which vascular signals remain consistent while noise varies. HA2HA was trained using in-vivo contrast-free pig kidney data and validated across diverse datasets, including contrast-free and contrast-enhanced data from pig kidneys, as well as human liver and kidney. An improvement exceeding 15 dB in both contrast-to-noise ratio (CNR) and SNR was observed, indicating a substantial enhancement in image quality. In addition to power Doppler imaging, denoising directly in the RF domain is also beneficial for other downstream processing such as color Doppler imaging (CDI). CDI results of human liver derived from the HA2HA-denoised signals exhibited improved microvascular flow visualization, with a suppressed noisy background. HA2HA offers a label-free, generalizable, and clinically applicable solution for robust vascular imaging in both contrast-free and contrast-enhanced UMI.