Examination of Breast Cancer Histology Images with MobileNet and NASNet Models
V. Rajinikanth
Abstract
In women community, the Breast-Cancer (BC) causes a larger infection-rate and death-rate. When the BC is detected in its early phase, it can be completely treated with necessary clinical procedure. Early detection and confirmation of BC involves in various imaging procedures including histology data. Manual examination of histology images are quite complex and hence, this research proposed deep-learning (DL) based tool and to improve its outcome, image pre-processing is executed. The various phases of this DL-tool consist the following phases; image collection and size modification, features extraction with DL-model and implementing classification with SoftMax, identification of best DL-models for generating fused-features-vector (FFV) using 50% reduction and serial fusion, and performance verification using classification and 5-fold cross validation. This research demonstrated the performance with MobileNet-variants and NASNet-variants using 5000 images of benign/malignant class. The result of this work confirms that the FFV-based analysis provides >97%.