Breast Cancer Detection Model Using Multi-Variable Clinical Data and Advanced Image Processing Techniques
Keywords:
Breast most cancers Prediction, Picture Processing, Convolutional Neural Network (CNN), data evaluation, Malignant and Benign Tissue type, Early Detection, Artificial Intelligence in Healthcare, Scientific Imaging, Life-style-primarily based Screening, Public Awareness and Perception SurveyAbstract
Breast cancer remains a leading cause of mortality among women worldwide, and early detection continues to be the most effective approach for improving survival outcomes. Despite its importance, traditional screening techniques often encounter obstacles such as high financial costs, limited availability in underserved regions, insufficient numbers of trained radiologists, and inconsistencies in diagnostic interpretation. To address these limitations, this study proposes a three-part AI-driven breast cancer prediction system that integrates self-assessment, structured diagnostic features, and medical imaging to form a comprehensive early-screening (Julien Florkin, n.d.) support model. The first component is a general risk evaluation questionnaire incorporating personal, genetic, hormonal, and lifestyle variables to facilitate preliminary awareness and risk stratification, particularly in resource-constrained environments. The second module employs structured clinical features—such as mean radius, texture, compactness, concavity, and symmetry—from the Breast Cancer Wisconsin (Diagnostic) Dataset to train a machine learning classifier capable of identifying malignancy based on established diagnostic indicators. The third component utilizes deep learning with transfer learning, leveraging the VGG16 architecture to classify ultrasound images from the BUSI dataset into normal, benign, or malignant categories. The imaging-based deep learning model demonstrated steady learning advancement, with training accuracy improving from approximately 46% to over 80%, while validation accuracy consistently remained between 75% and 77%. The continuous decline in both training and validation loss—supported by early stopping—indicated effective generalization and reduced overfitting. The structured-data model also achieved high predictive accuracy, reaffirming the discriminative power of clinical features in malignancy detection. Furthermore, a feasibility and user-feedback survey conducted as part of this research revealed strong acceptance of AI-assisted screening tools, emphasising perceived usefulness, ease of use, and their potential role in supporting healthcare professionals. In conclusion, the proposed multimodal framework illustrates the promise of integrating AI-based risk assessment, conventional machine learning, and deep learning to deliver accessible, affordable, and reliable early breast cancer detection. This unified machine lays a strong foundation for the future development of deployable screening answers and clinical selection-guide packages.
