Multimodal Analysis of Hereditary Breast and Ovarian Cancer Syndrome using Deep Learning techniques

Hereditary Breast and Ovarian Cancer (HBOC) Syndrome is an autosomal dominant genetic condition, necessitating early and accurate detection for effective management. This study aims to develop a deep learning framework to analyze the image and gene data for HBOC syndrome. In this study, the breast ultrasound image dataset is obtained from Kaggle and the real time ovarian ultrasound images are obtained from hospitals. The gene sequence was obtained from NCBI Gene and NCBI ClinVar. The images are preprocessed and features are extracted from the images using transfer learning technique. Based on the comparative analysis, InceptionV3 is found to be the most effective model for feature extraction. An ensemble model based on a voting scheme for classification of breast images into normal, malignant and benign based on the features, is proposed. Similarly, for the classification of ovarian images into normal and abnormal, another ensemble model is proposed. The proposed model demonstrates 98% accuracy for ovarian image classification and 94% acccuracy in breast image classification. The gene sequences are preprocessed and numerical features are extracted. For the classification of normal and mutated genes, an ensemble model of classifiers is used. This prediction is used as a preventive strategy through periodic screening of cancer. The outcome of this study would provide a holistic approach to cancer risk assessment, providing a better diagnostic accuracy.