Benign/Malignant Breast Cancer Detection in Histopathology Image using Fused Deep-Learning Features

Abnormal cell growth in organs is recognized as cancer, one of the most severe and life-threatening diseases. Timely detection of cancer is essential to treat the illness using recommended treatment procedures. This research work considered the Breast Cancer (BC) for the investigation. It is one of the common cancers in women and clinical level diagnosis is performed using the medical-image guided methods. Histopathology-image based examination of the BC is essential for detecting its harshness. This research aims to develop Lightweight Deep-learning (LD) approach to classify the images into benign/malignant class. The stages of this scheme includes; image collection and resizing, feature extraction using the LD approach, feature reduction using 50% dropout and serial features generation, and classification and 3-fold cross validation. This investigation considered the BreakHis benchmark database with 40X and 100X magnification for the study. In this work, the LD methods, like MobileNet and NASNet are considered to develop the proposed examination scheme. BC detection is executed using the individual-features (IF), and fused-features (FF) using SoftMax classifier. The experimental outcome of this study confirms that the FF based classification provides 100% detection accuracy on both 40X and 100X magnified images.