Find the papers that actually matter

Breast cancer remains the most prevalent malignancy among women worldwide. The timely detection of this cancer type is critical for improving survival outcomes. Despite advancements, mammogram classification using deep learning strategies still faces challenges. These include inter-view feature inconsistency, loss of diagnostic details, and limited interpretability. In order to address these issues, MammoFusion-Net, a dual-branch deep learning framework, is proposed for mammogram-based breast cancer classification. Using residual convolutional streams, the framework processes craniocaudal (CC) and mediolateral oblique (MLO) views independently. This supports preservation of view-specific anatomical information. In the proposed framework, a Gates Cross-View Fusion mechanism adaptively integrates features across views. As a result of experimental analysis, the proposed framework achieved 92.116 % (VinDr-Mammo dataset) and 95.556 % (INBreast dataset) of improved classification performance.•Employs a dual-branch architecture to independently process CC and MLO views using residual convolutional streams.•Integrates Gated Cross-View Fusion and attention mechanisms adaptively and refines multi-view features for stronger discrimination.•Demonstrates the explainability of the model through Grad-CAM visualizations that highlight lesion-relevant regions.

Neoadjuvant systemic therapy has emerged as a strategy to improve outcomes in high-risk localized genitourinary malignancies. In bladder cancer, neoadjuvant cisplatin-based chemotherapy with or without immunotherapy is standard of care, with pathologic complete response (pCR) serving as a validated surrogate for survival after chemotherapy and a promising potential surrogate for survival after other neoadjuvant treatments like immunotherapy. Enfortumab vedotin and immune checkpoint inhibitors have expanded treatment options, with ongoing studies evaluating novel adjuvant approaches tailored to patients' postoperative circulating tumor DNA. In renal cell carcinoma (RCC), the field is nascent, no approved neoadjuvant regimens exist, and treatment outside clinical trials is not recommended. Early trials of tyrosine kinase inhibitor monotherapy showed limited pathologic responses, while immune checkpoint inhibitor-based combinations have demonstrated feasibility, safety, and the capacity to induce pathologic responses including pCR. Critical gaps remain in both diseases. In RCC, standardized pathologic response criteria are lacking, and surrogacy between pathologic end points and long-term outcomes has not been established. In bladder cancer, optimal post-pCR management and integration of novel agents require further study. Emerging technologies, particularly artificial intelligence (AI)-driven digital pathology, offer potential to enhance diagnosis, refine prognostic stratification, and predict treatment response, although prospective validation across diverse populations is needed. This chapter examines neoadjuvant therapy and pathologic response assessment in RCC and bladder cancer, explores pathologic biomarker development including AI applications, and highlights future directions to optimize therapeutic sequencing and outcomes.

Acute lymphoblastic leukemia (ALL) is a hematological malignancy characterized by the rapid proliferation of immature white blood cells in the bone marrow. Early and accurate diagnosis is essential for improving clinical outcomes; however, distinguishing between lymphocytes and lymphoblasts poses significant challenges owing to their subtle morphological similarities. Traditional manual diagnostic methods, which rely on expert evaluations, are inherently time-consuming and subject to human error. In recent years, machine learning and deep learning approaches have emerged as promising tools for automating and enhancing diagnostic processes. This review systematically examines state-of-the-art traditional and deep learning techniques applied for ALL detection and classification. We provide a comprehensive analysis of various methodologies, including supervised machine learning algorithms and advanced deep learning architectures, with a focus on critical stages such as image preprocessing, feature extraction, and blast cell quantification. Furthermore, we discuss the performance metrics and accuracy benchmarks, highlighting the potential of these techniques to match or exceed human diagnostic capabilities. The review concludes with a discussion of the current challenges, recent developments, and future directions in the application of artificial intelligence for ALL diagnosis, underscoring the need for continued innovation to meet emerging clinical demands.

Prostate cancer represents the second most common malignancy among men globally, necessitating accurate diagnostic methodologies for optimal patient outcomes. Convolutional neural networks (CNNs), a core deep learning methodology, have emerged as transformative technologies for automated prostate cancer detection, classification, and segmentation across multiple imaging modalities. A systematic review following PRISMA guidelines was conducted across Web of Science, Scopus, and PubMed databases (January 2020-December 2025). CNN-based classification architectures were analyzed across ResNet, Vision Transformer, DenseNet, Xception, ConvNeXT, and Swin Transformer implementations, with comparative evaluation of accuracy and transfer learning performance. Object detection and segmentation approaches were examined across U-Net variants, R-CNN family algorithms, and YOLO-based implementations. Hyperparameter optimization strategies were assessed. Explainable AI methodologies including SHAP, Grad-CAM, DiCE, and LIME were evaluated for clinical interpretability and spatial localization accuracy. Analysis of 320 publications revealed peak research activity in 2024 (63 publications, 19.7%). The United States led with 58 publications (18.1%), followed by China with 55 (17.2%). Multiparametric MRI constituted the primary imaging modality (42.5%), followed by histopathology (28.1%), ultrasound (14.1%), and PET imaging (9.4%). Vision Transformer models demonstrated the highest classification accuracy among evaluated architectures, while U-Net variants dominated segmentation applications with consistently high Dice coefficients. SHAP emerged as the most frequently adopted explainability method across the reviewed studies. CNN-based prostate cancer detection, classification, and segmentation demonstrate promise for improving diagnostic accuracy and clinical workflow efficiency, though challenges in dataset standardization, regulatory compliance, and clinical integration remain to be addressed.

Urinary proteins are promising non-invasive biomarkers, but their low abundance and wide dynamic range make detection challenging. This study presents UriPred, a computational tool that integrates machine learning (ML), BLAST, and Motif-EmeRging and Classes-Identification (MERCI) to predict urinary proteins and facilitate the identification of liver cancer (LC) biomarkers. A dataset of 10588 urinary and non-urinary proteins was curated, from which two feature types were generated: 10074 compositional and 20 evolutionary features. Seven feature selection methods were applied to compositional features, and 11 ML algorithms were trained on different feature sets. Evolutionary features achieved the highest training performance (AUC 0.79, accuracy 71.99 %), whereas amino acid composition (AAC) with 20 features achieved identical validation AUC (0.74) and comparable accuracy while being computationally less expensive and consistently selected. The ML-AAC model was therefore chosen as the final model. This optimal model was integrated with BLAST and MERCI to create UriPred, which reduced false positives from 34.59 % (ML) to 3.12 % (hybrid) on the validation dataset and from 5.8 % (ML) to zero (hybrid) on an external dataset. Using UriPred, 53 LC differentially expressed protein-coding genes were predicted as urinary proteins. Protein-protein interaction analysis, AUROC evaluation (AUC > 0.80), survival analysis, and cross-verification of urine detectability with the Human Protein Atlas and Human Urine PeptideAtlas databases identified five proteins (KIF23, COL15A1, CTHRC1, MMP9, and SPP1) as potential LC biomarkers. UriPred efficiently predicts urinary proteins using AAC features and enables biomarker discovery for LC. The tool is publicly available at https://github.com/Dahrii-Paul/UriPred.

Glioblastoma multiforme (GBM) is the most aggressive primary brain tumor of the central nervous system and remains associated with poor prognosis. Although treatment strategies have improved, the blood-brain barrier (BBB) continues to impede effective drug delivery. Here, we implemented an integrative computational framework combining transcriptomic analysis, network biology, machine learning (ML), and structure-based validation to prioritize potential FDA-approved drug candidates for GBM. Differential gene expression analysis and network-based topological ranking identified TOP2A, KIF20A, and KIF11 as highly connected hub genes. Bioactivity data for TOP2A and KIF11 were curated from ChEMBL, rigorously filtered, and encoded using PaDEL fingerprints. The best-performing regression model for KIF11 was used to screen an FDA-approved drug library, identifying compounds with predicted potency (pIC50 ≥ 6.5) and predicted BBB permeability. Selected candidates were further evaluated using molecular docking and 500 ns all-atom molecular dynamics simulations with MM-GBSA calculations to assess structural stability and relative binding energetics. Among four prioritized compounds, Ponatinib demonstrated the most favorable binding free energy, while Pimavanserin exhibited stable conformational behavior during simulation. These findings provide an in-silico prioritization framework for potential KIF11-targeting compounds in GBM. Experimental validation in relevant cellular and in vivo models will be required to determine biological and therapeutic relevance.

Breast cancer is a life-threatening disease that is very common among women in the world. The early and correct diagnosis is necessary to enhance the rate of survival and treatment. The conventional techniques such as mammography, ultrasound and MRI usually fail to differentiate between the benign and malignancy tumors in the presence of limited resources, particularly in the resource-poor environments. This paper presents a new methodology of breast cancer classification, Shape-Aware Angular Feature Learning (SAAFL), that merges machine and deep learning methods. We propose a method of Speckle-Reducing Anisotropic Diffusion (SRAD) filters to improve the quality of ultrasound images by reducing the number of speckle noise and maintaining the edges of the tumors. Our segmentation approach is robust, which is RBBSAM-RSF, using which we detect tumors automatically and then process the data with Angular Feature (AF) analysis in order to determine the specific features of the lesions. The hierarchical classification system combines the AF extraction and classifiers like Support Vector Machines (SVM) and Backpropagation Artificial Neural Networks (BPANN) to enhance the accuracy of the diagnostic. On 1293 breast ultrasound (BUS) images, our model manages to attain 95.38 % accuracy, which is better than the traditional texture-based and morphological models. This reduces false positives and unnecessary biopsies, making it suitable for near-real-time deployment in low-resource clinical settings without specialized hardware. By selectively integrating deep learning-assisted segmentation with shape-aware angular feature analysis and supervised machine learning, SAAFL advances noninvasive and interpretable breast cancer diagnostics.

Optical biopsy enables minimally invasive, quantitative tissue assessment, yet clinically useful implementations require rapid and objective decision-making from compact sensors. We present a refractive-index (RI) driven classification framework based on reflection spectra acquired with an extrinsic fiber-optic Fabry-Pérot interferometric cavity (280 μm) over the biologically relevant RI range 1.33-1.42. Three proxy classes ("healthy", "HCC-like", "metastatic") were defined using literature-guided RI windows for liver tissue, and measurements were performed on certified reference liquids. As a physics-only reference, RI was estimated analytically from fringe periodicity in the wavenumber domain, achieving 0.70 accuracy and 0.48 macro-F1. To enhance discrimination, we engineered 62 spectral descriptors capturing fringe spacing (RI-related), fringe visibility, and spectral-shape cues, and trained tree-ensemble and SVM models together with an interpretable GA-optimized fuzzy expert system. On a held-out test set, tree ensembles reached macro-F1 = 1.00, while SVM and the fuzzy system achieved 0.96 and 0.97, respectively. Feature attribution identified RI as the dominant discriminative signal, with visibility-related metrics improving robustness near the HCC-like boundary. These results demonstrate that ML-augmented fiber-optic interferometry can deliver accurate and explainable diagnostic signatures, supporting the translational potential of RI-based optical biopsy.

Approximately 30-50% of Papillary thyroid carcinoma (PTC) patients develop cervical lymph nodes (LNs) metastasis, significantly increasing the risk of disease recurrence and impacting long-term outcomes. We introduced an open-access multicenter lymph node ultrasound image database (LymphUs) specifically designed to advance research in LN assessment for PTC. Ultrasound imaging was performed on PTC patients at two independent clinical centers using standardized acquisition protocols. Experienced radiologists at each center documented sixteen semantic features for each LN. All LNs were annotated with segmentation masks serving as ground truth, and classification into benign or malignant categories was confirmed by fine needle aspiration biopsy results. The LymphUs comprises ultrasound images with segmentation masks from 338 PTC patients with suspected LN metastasis, divided into two center-specific cohorts: 180 patients (81 malignant, 99 benign) and 158 patients (82 malignant, 76 benign). The complete dataset, including semantic features and expert annotations, is freely accessible for research purposes. The LymphUs bridges a critical gap in medical imaging resources by providing a large-scale, multicenter ultrasound database for cervical LN assessment in PTC, supporting diagnostic algorithms, standardized reporting systems, and artificial intelligence applications to enhance preoperative LN staging and treatment planning.

Prostatic adenocarcinoma (PRAD) is among the most common malignancies in men and is characterized by considerable genetic and epigenetic heterogeneity. Despite advances in diagnosis and treatment, options for advanced and refractory prostate cancer remain limited, which adversely affects patient prognosis. This project aims to identify diagnostic and prognostic biomarkers associated with programmed cell death (PCD) mechanisms in prostate cancer and to reposition existing drugs that target these biomarkers. Using RNA-seq and clinical data from The Cancer Genome Atlas (TCGA), differential gene expression, ROC curve, and survival analyses identified six candidate biomarkers with strong diagnostic and prognostic significance. Three small molecules - DL-PDMP, clobetasol propionate, and metoclopramide hydrochloride - capable of reversing gene expression profiles were selected for in vitro assays in a drug repositioning analysis conducted using the L1000CDS2 platform. Among these, DL-PDMP was prioritized because of its low IC50 value and low toxicity to normal prostate epithelial cells. Furthermore, DL-PDMP has been shown to induce apoptosis and suppress colony formation. These findings suggest that targeting PCD-associated biomarkers is a promising strategy for prostate cancer treatment, making DL-PDMP a strong candidate for further preclinical studies.