A Comparison Based Breast Cancer High Microscopy Image Classification Using Pre-trained Models
Nidhi Ojha, Ashwani Kumar · 2020
Breast Cancer is most common but difficult to diagnose at an early stage, commonly found in women but there are some cases of men also. The early tedious way of treating breast cancer was quite time-taking as well as causes a dispute between the pathologists based on their analysis. Due to this, pathologists were not able to find out the cancerous tissues on time. Ordinarily, the malignancy shapes in either of the breast. Deep learning has shown much improvement in the diagnosis of Breast cancer, we propose a new technique to diagnose breast cancer tissue by the processing of the biopsy or histopathology images. These images are of large dimensions. Passing these images will require a high computational GPU and it is not possible to use. So we break it into different smaller sizes. Here, we are using H&E staining to make the images more clear to increase the chances of detecting the cancerous tissue. The image extracted from histopathology images utilizing the pre-trained CNN models VGG 19, Inception V3 and ResNet 152 with various sizes of crops can precisely recognize typical and malignant growth tests. The proposed model is state of the heart in terms of accuracy than all previously reported work, using CNN and LightGBM.