Neural Architecture Search for Histopathological Image Classification

The accurate classification of histopathology images is important in medical diagnosis processes. In recent years, Neural Architecture Search (NAS) methods have been developed to optimize the model selection process, one of the important problems in this field, by enabling the automatic design of deep learning models. This study proposes a new neural architecture search method for histopathology image classification using the Opposition-Based Differential Evolution (ODE) algorithm. The proposed method is based on the PBC-NAS architecture and includes a search strategy developed to solve plateau problems. The performance of the proposed method is evaluated on the Enteroscope Biopsy Histopathological H&E (EBHI) image dataset. In the experimental studies, the proposed method is compared with deep learning models such as ResNet, MobileNet, and DenseNet, which are widely used in the literature in terms of accuracy and processing time. The proposed method achieves highly competitive results (0.5 points difference) with up to 5 times less computation time, which is crucial for medical image analysis.