Analyzing Histopathological Images for Cancer Prediction using Human Centric Learning Approaches

Abstract

Examining Histopathological images are a substantial approach for earlier cancer prediction in clinical analysis. However, the examination encounters some inefficiency; therefore, the cancer prediction process is depicted as a significant issue in medical imaging analysis. To simulate the prediction accuracy and to diminish the expert's decision-making complexity, this work proposes a novel feature extraction and selection of histopathological images by integrating deep learning and machine learning approaches. Initially, the provided input samples are pre-processed via dimensionality reduction, RGB colour analysis, and image transformation. Then, the features are extracted with the pre-trained network model like AlexNet, GoogleNet, Inception V3, and ResNet 50. Next, feature selection is done with Recursive Feature Elimination (RFE) to enhance and boost the system performance and eliminate over-fitting or under-fitting issues. The proposed model is evaluated with the key evaluation parameters like accuracy, precision and recall. At last, a non-linear Support Vector Machine ($nl-SVM$) is trained to fuse the related features and to enhance the performance outcomes. Here, an online available dataset for histology image-based cancer analysis is adopted. The observation proves that the anticipated model gives promising outcomes and better results than various prevailing approaches.