International Journal of Science, Engineering and TechnologyAuthors: Dr. Pritesh Patil, Abhiraj Bondre, Gauri Dighe, Kiran Mangde
Abstract: Breast cancer is one of the most common and life-threatening diseases affecting women worldwide, with approximately 2.3 million new diagnoses reported each year. Despite significant improvements in survival rates, early and accurate detection remains a major clinical challenge, especially in resource-constrained healthcare settings where radiologist availability and turnaround times can directly influence patient outcomes. In many hospitals, a single radiologist may be responsible for reviewing hundreds of MRI scans per day under time pressure — a situation that almost inevitably leads to some degree of inconsistency and missed findings. This paper introduces Women Wellness, a fully integrated diagnostic platform designed to tackle these challenges by combining deep learning-based image classification, OpenCV-powered video preprocessing, cloud-hosted parallel inference, and automated clinical report generation within a single deployable system. The platform accepts breast MRI video sequences as input, extracts individual frames on the client side using the HTML5 Canvas API, preprocesses them through OpenCV.js, and routes them through a fine-tuned Convolutional Neural Network hosted as Python Flask microservices. Classification results — spanning normal, benign, and malignant categories — are aggregated using a confidence-weighted voting scheme, and the entire pipeline culminates in a structured, multi-page PDF report generated automatically via jsPDF. In experiments conducted on a curated MRI dataset, the system achieved overall classification accuracy between 92% and 95%, with malignant case sensitivity reaching 96.3%. The complete analysis and report pipeline completes in 45 to 60 seconds per study, compared to roughly 15 to 20 minutes for conventional manual review. Grad-CAM attention maps are embedded directly into each report, enabling radiologists to visually verify which image regions most influenced each classification decision rather than simply taking the model's word for it.
