AN AI-ASSISTED CLINICAL DECISION SUPPORT SYSTEM FOR EARLY BREAST CANCER DETECTION AND RISK STRATIFICATION USING CELLULAR MORPHOMETRICS

Abstract

Breast cancer remains one of the leading causes of cancer-related mortality worldwide, with delayed diagnosis significantly reducing treatment effectiveness and patient survival rates. Conventional diagnostic workflows involving pathological interpretation and radiological assessment are often time-intensive and subject to inter observer variability. Recent advances in Artificial Intelligence (AI) and machine learning have demonstrated substantial potential in enhancing diagnostic accuracy, improving clinical workflow efficiency, and enabling precision-based healthcare. This thesis proposes an AI-assisted Clinical Decision Support System (AI-CDSS) for early breast cancer detection and risk stratification using structured cellular morphometric features. The study utilizes the Breast Cancer Wisconsin (Diagnostic) dataset consisting of 569 patient samples and 30 quantitatively extracted cytological features derived from digitized fine needle aspiration images. Four machine learning algorithms, namely Logistic Regression, Random Forest, Gradient Boosting, and Multi-Layer Perceptron Neural Network, were developed and comparatively evaluated for binary tumor classification into benign and malignant categories. Data preprocessing involved normalization using standard scaling and dataset partitioning into training and testing subsets for unbiased performance evaluation. Among the evaluated models, Logistic Regression demonstrated superior clinical applicability with an AUROC score of 0.9960, sensitivity of 0.9286, and specificity of 0.9861. Explainable AI analysis further identified “worst area” and “worst concave points” as the most influential predictive features associated with malignancy. A probability-based risk stratification framework was also developed to categorize patients into low-risk, intermediate-risk, and high-risk groups. The proposed AI-CDSS offers a transparent, computationally efficient, and clinically interpretable framework for supporting oncologists in early breast cancer diagnosis and patient triaging. The study highlights the potential integration of explainable AI systems into modern oncology workflows while emphasizing the importance of maintaining physician oversight for clinically ambiguous cases.