Comparison of CNNs for Lung Biopsy Images Classification
Daria Hlavcheva, Vladyslav Yaloveha, Andrii Podorozhniak, Heorhii Kuchuk
Abstract
Deep learning approaches are widely used in the processing of medical images, including histopathological images for cancer diagnosis. Therefore, the scientific and practical problem of automation of biopsy image analysis using convolutional neural networks is considered in the paper. The LC25000 dataset was used to compare the classification accuracy of different CNN architectures. To analyze the impact of image size, two more datasets were created from the initial dataset by slicing of the images. The correlation between the complexity of CNN structure, size of the images, and the resulted accuracy on test data was obtained. Results were compared with related researches on the LC25000 dataset. The theory of deep learning neural networks and mathematical statistics methods are used.