Design and Development of AI based Approach for Histopathology Cancer Screening and Identification
Cancer is a heterogeneous disorder comprising various types and sub-types. Early detection, screening, and diagnosis of cancer types are necessary for facilitating cancer research in early diagnosis, management, and the evolution of successful therapies. Existing methodologies were only able to classify and diagnose a single variety of cancer based on a homogeneous dataset but more focused on predicting patient survivability then cure. This research defines a machine learning-based methodology to develop an universal approach in diagnosis, detection, symptoms-based prediction, and screening of histopathology cancer, their types, and sub types using a heterogeneous dataset based on images and scans. In this architecture, we use VGG-19 based 3D-Convolutional Neural Network for deep feature extraction and later perform regression using a random forest algorithm. We create a heterogeneous dataset consisting of results from laboratory tests, imaging tests and biopsy reports, not only relying on clinical images. Initially, we categorize tumors and lesions as benign or malignant and classify the malignant lesions into their sub-types, detecting their severity and growth rate. Our system is designed to predict risk at multiple time-points, leverage optional risk factors if they are available and produce predictions that are consistent across mammography machines. We found the classification accuracy for categorizing tumors as cancerous to be 95% whereas the accuracy for classification of malignant lesions into their sub-types to be 94%.