Prostate cancer (PCa) is the second most common cancer type in men, and the need for a non-invasive PCa diagnostic and stratification tool can now be met through the quantified analysis of multiparametric MRI. Early diagnosis of clinically significant prostate cancer affects patient management. Radiomics analysis can quantify imaging information in a region of interest (ROI) that is not visible to clinicians, thereby improving cancer detection. Radiomic based models have been proposed for the diagnosis of csPCa, mainly focusing on the evaluation of lesions, a process that is both challenging and time consuming. On the other hand, automation of whole gland segmentation is achieved using deep learning algorithms and radiomic-based model considering whole prostate gland have shown promising results. In this study we investigate whether radiomics from different parts of the gland affect classification performance. T2 weighted and diffusion weighted images (DWI), of different b values, from 80 patients were analyzed. Radiomics were extracted from the gland, the transition and periphery zone and areas with predefines widths. Recursive feature elimination together with a voting strategy was used for feature selection and two machine learning models were trained and tested. The results revealed that the best diagnostic accuracy of 81.9% (±7.8) was succeeded using the radiomics from the transition and periphery zone whereas, DWI of high b value (800) is the most informative imaging modality. These findings highlight the need for the development of an accurate segmentation algorithm which can improve the development of accurate and robust models for the diagnosis of csPCa.Clinical Relevance- This work can guide radiomic research for the development of accurate and robust models for the diagnosis of csPCa assisting clinicians in patient evaluation and management.

Gastric cancer (GC) remains a significant global health issue, and convolutional neural networks (CNNs) have shown their high potential for detecting precancerous gastrointestinal (GI) conditions on endoscopic images [1] [2]. Despite the need for high resolution to capture the complexity of GI tissue patterns, the impact of endoscopic image resolution on the performance of these models remains underexplored. This study investigates how different image resolutions affect CNNs classification of intestinal metaplasia (IM) using two datasets with different resolutions and imaging modalities. Our results reveal that the often adopted input resolution of 224×224 pixels does not provide optimal performance for detecting IM, even when using transfer learning from networks pre-trained on images with this resolution. Higher resolutions, such as 512×512, consistently outperform 224 × 224, with notable improvements in F1-scores (e.g., InceptionV3: 94.46% at 512 × 512 vs. 91.49% at 224 × 224). Additionally, our findings indicate that model performance is constrained by the original image quality, underscoring the critical importance of maintaining the higher original image resolutions and quality provided by endoscopes during clinical exams, for the purposes of training and testing CNNs for gastric cancer management.Clinical Relevance- This research highlights the importance of image quality, particularly when endoscopes capture lower-resolution images. Understanding how image resolution impacts diagnostic accuracy can guide clinicians in improving imaging techniques and employing Artificial Intelligence-driven tools effectively for more accurate GC detection and better patient outcomes.

Early detection of non-small cell lung cancer (NSCLC) is critical for improving patient outcomes, and novel approaches are needed to facilitate early diagnosis. In this study, we explore the use of automatic cough analysis as a pre-screening tool for distinguishing between NSCLC patients and healthy controls. Cough audio recordings were prospectively acquired from a total of 227 subjects, divided into NSCLC patients and healthy controls. The recordings were analyzed using machine learning techniques, such as support vector machine (SVM) and XGBoost, as well as deep learning approaches, specifically convolutional neural networks (CNN) and transfer learning with VGG16. To enhance the interpretability of the machine learning model, we utilized Shapley Additive Explanations (SHAP). The fairness of the models across demographic groups was assessed by comparing the performance of the best model across different age groups ( 58y and >58y) and gender using the equalized odds difference≤ on the test set. The results demonstrate that CNN achieves the best performance, with an accuracy of 0.83 on the test set. Nevertheless, SVM achieves slightly lower performances (accuracy of 0.76 in validation and 0.78 in the test set), making it suitable in contexts with low computational power. The use of SHAP for SVM interpretation further enhances model transparency, making it more trustworthy for clinical applications. Fairness analysis shows slightly higher disparity across age (0.15) than gender (0.09) on the test set. Therefore, to strengthen our findings' reliability, a larger, more diverse, and unbiased dataset is needed-particularly including individuals at risk of NSCLC and those in early disease stages.Clinical relevance- This study highlights the promise of integrating automatic cough analysis with machine learning techniques for improving lung cancer screening methods.

This paper addresses the challenge of weakly supervised learning for breast cancer detection in mammography by introducing an Enhanced Embedded Space MI-Net model with deep supervision. The framework integrated adaptive patch creation, convolution feature extraction, and pooling methods -max, mean, log-sum-expo, attention, and gated attention pooling - evaluated in three MIL models, Instance Space mi-Net, Embedded Space MI-Net and Enhanced Embedded Space MI-Net. A key contribution is the incorporation of deep supervision, improving feature learning across network layers and enhancing bag-level classification performance. Experimental results on the CBIS / DDSM dataset demonstrate that the Enhanced MI-Net model achieves the highest AUC of 86% with attention pooling. This work addresses the gap in leveraging MIL techniques for high-resolution medical imaging without requiring detailed annotations, offering a robust and scalable solution for breast cancer detection.Clinical Relevance-This study highlights the potential of MIL-based models with attention pooling to accurately detect breast cancer in mammographic images without requiring detailed ROI annotations, offering a scalable and efficient diagnostic tool for clinical practice.

Brain metastases (BM), along with primary central nervous system lymphomas and glioblastomas, represent the majority of malignant brain tumors encountered in clinical neuro-oncology, driving a need for advanced imaging techniques and post-processing methods to improve their characterization and treatment monitoring. In particular, stereotactic radiosurgery (SRS), a cornerstone treatment for BM, delivers high-dose, focused radiation (>20 Gy) to target lesions with minimal impact on surrounding tissues. Despite its efficacy, radiation-induced effects such as early radiation effects (ERE) and adverse radiation effects (ARE) complicate diagnosis and management, with ARE occurring in up to 30% of patients, often presenting as ring-enhancing T2/FLAIR hyperintensities. To address these challenges, we aimed to compare standard radiomics-based machine learning approaches with pretrained generative models for assessing ERE in BM lesions. A cohort of 21 patients for a total of 35 lesions (17 treatment-naïve and 18 post-SRS +/- combination therapy) who underwent multiparametric ¹⁸F-FPIA PET/MRI was analyzed. The study investigated: 1) Multiparametric analysis of PET and MRI diffusion/perfusion parameters (ADC, K^trans, CBF, K₁, v_t); 2) MRI-based radiomics; 3) static PET radiomics; 4) Dynomics; 5) a combination of PET and MRI radiomics; and 6) low-level embeddings from a pretrained generative diffusion model applied to full T1, static PET, and their combination. Using manually contoured lesion masks for analyses 1-5 and lesion-free embeddings for analysis 6, multiple classifiers (SVM, XGBoost, Linear regressor) were applied after feature standardization and principal component analysis (retaining 90% variance). Fivefold cross-validation demonstrated comparable performances across radiomic approaches (Accuracy: 71.95±0.05%, AUC: 0.72±0.05%), while the pretrained generative model achieved significantly higher performance (Accuracy: 83.82±0.01%, AUC: 0.83±0.01%) without requiring lesion segmentation in assessing ERE in BM lesions.Clinical Relevance-This study shows the potential of generative models to streamline and enhance the assessment of early radiation effects in parenchymal metastatic lesions without need of lesion segmentation.

Pancreatic cancer (PC) is a major global public health problem. Identifying potential biomarkers and individual patient characteristics that contribute to early PC detection is critical in clinical practice. The present study employed a comprehensive data-driven pipeline to address this important issue in a subset of patients diagnosed with Pancreatic Ductal Adenocarcinoma (PDAC) along non-oncologic samples. Putative extracellular vesicle (EV) characteristics in combination with clinical and laboratory features served as potential predictors of patient risk stratification and PDAC diagnosis. The machine learning (ML)-based pipeline entailed the appropriate steps for unbiased model training and validation. The two groups of samples were discriminating against a high degree of accuracy (= 0.96) with balanced sensitivity (=0.95) and specificity (=0.97) rates. Both EV-based variables and biochemical characteristics emerged as important predictors of PDAC diagnosis.Clinical Relevance-Minimally invasive technologies based on extracellular vesicles (EVs) along with adaptive learning methodologies could provide new directions and solutions with increased efficiency in PC clinical diagnosis.

This study explores the optimization of hyperthermia-mediated drug delivery using thermo-sensitive liposomes (TSLs) for treating hepatocellular carcinoma (HCC). By employing a Multi-Objective Genetic Algorithm (MOGA), the research aims to maximize tumor cell kill rates while minimizing thermal damage to tissues. The mathematical models used include the Pennes' bioheat equation and drug diffusion equations, integrated into finite element simulations. The optimization process balances critical parameters which drive both heating protocol and drug release mechanisms, resulting in improved therapeutic outcomes. This innovative approach addresses the challenges of effective TSL-mediated chemotherapy, providing a promising pathway for enhancing clinical treatments of HCC.Clinical Relevance- This study is significant for clinicians as it proposes a computational method to obtain optimized input protocols for hyperthermia-mediated drug delivery in HCC. By fine-tuning treatment parameters, the approach aims to increase drug efficacy while reducing side effects, offering a more targeted and efficient alternative to conventional chemotherapy.

Dynamic Contrast-enhanced MRI (DCE-MRI) synthesis is crucial for improving prostate cancer diagnosis by reducing reliance on contrast agents. Previous deep learning approaches use single-stage architectures, limiting the progressive refinement of perfusion features essential for preserving diagnostically relevant information. We propose MPM-DCE, a multi-stage GAN-based framework that synthesizes early and late response DCE-MRI images using multimodal inputs. The proposed encoder-decoder generator progressively refines features by expanding spatial context and exchanging information between stages sequentially and laterally via an Adaptive Inter-Cascade Feature fusion (AICF) blocks to learn low-level perfusion details within the anatomy. Extensive evaluations on the ProstateX dataset demonstrate that MPM-DCE (i) achieves state-of-the-art performance, surpassing existing approaches such as DCE-Former with +0.31 dB and +0.65 dB PSNR, +0.01 and +0.03 SSIM improvements, and -0.02 and -0.02 MAE reduction for early and late responses respectively and (ii) highlights the significance of multistage progressive feature refinement and inter-stage attention mechanisms for improved DCE-MRI synthesis.Clinical relevance - The proposed method achieves statistically significant improvement (Wilcoxon signed-rank test, p < 0.05) in DCE-MRI synthesis, preserving perfusion details while reducing reliance on gadolinium contrast agent, potentially improving prostate cancer diagnostics and minimizing unnecessary biopsies.

We present a proof-of-concept for the extension of the Brain Imaging Data Structure (BIDS) to accommodate Computed Tomography (CT) data. With the growing volume of CT imaging across various medical fields, including neuroradiology and thoracic imaging, the need for data standardization is increasingly critical, especially in the context of artificial intelligence (AI) tools for medicine. This study demonstrates the conversion of OASIS-3 and National Lung Screening Trial (NLST) datasets into BIDS format and the development of a BIDS App for lung cancer risk prediction using the Sybil AI tool. The resulting framework promotes interoperable, accessible, and reusable data, fostering Open Science and enabling independent validation of AI models across diverse systems and datasets, ultimately addressing challenges like bias and overfitting in clinical settings.Clinical relevanceThis study enables the sharing and reuse of CT data within the research community, enhancing knowledge extraction and accelerating the development and validation of AI tools that can improve diagnostic accuracy and clinical decision-making across various medical fields.

Automatic classification of breast cancer (BrCan) histopathological images is crucial for aiding BrCan diagnosis. Convolutional neural networks often emphasize semantics, however, they face challenges like high computational cost, high memory usage, and difficulty in capturing multiscale features, making them less suitable for resource-constrained clinical applications. Many past researchers have proposed CNN-based deep learning model for BrCan classification. Though these models achieved good classification performance, they are heavy regarding parameter counts, FLOPS, and model size. This paper presents a lightweight model with performance comparable to state-of-the-art methods. It integrates Depthwise-Dilated-Multiscale-Pointwise (DDMP) blocks, Discrete Wavelet Transform (DWT), and Squeeze-and-Excitation (SE) blocks to capture low-level and high-level discriminative features. The DDMP blocks efficiently extract multiscale features using depthwise convolution, multiscale dilated convolutions, and pointwise convolutions. This is followed by dual-stream architecture combining the LL subband of DWT with max-pooling output. These features are then recalibrated using SE blocks to highlight the most significant features. The proposed model consists of two DDMP-SE blocks, followed by Global Average Pooling, dense layers with GELU activation, and a final soft-max layer for binary classification. An ablation study further highlights the impact of epochs, activation functions, and batch sizes. BreastHistoNet outperforms other baseline models in terms of model size (7.47 MB), parameter count (0.63 M), and FLOPS (6.50 G). Experimental results on BreaKHis dataset obtain 95.48% accuracy, 95.61% precision, 95.46% specificity, 95.46% recall, and 95.48% F1-score. The performance of BreastHistoNet offers high accuracy while maintaining low computational complexity and minimal memory usage makes it a valuable tool for accurate and efficient BrCan classification.