Experimental Study of an Active Learning-based Method for the Classification of Penile Cancer in Histopathological Images using Convolutional Neural Networks

Abstract

Penile cancer is a disease with a significant incidence in developing countries. One of the main diagnostic methods for this type of cancer is the histopathological analysis of tissues collected through biopsy, which is time-consuming. Deep Learning methods emerge as a promising alternative. However, conventional supervised model training is not always feasible because it generally requires a large amount of labeled data to achieve satisfactory results. Active Learning emerges as a potential solution, as it considers the selection of the most informative examples to be labeled by specialists. This work presents an Active Learning-based method for penile cancer classification in histopathological images as part of an experimental study. Tests were conducted with images at 100X magnification, and the results obtained are promising, with emphasis on the performance of the EfficientNet B0 network, which achieved an accuracy of 0.9399, F1-Score of 0.9473, sensitivity of 0.9543, and precision of 0.9413.

Ähnliche Arbeiten