Detection of Overlapping Cells in Histopathological Images with Deep Sparse Learning

Abstract

Nowadays, artificial intelligence is rapidly developing, and with increasing data volumes, the learning capacity of models is expanding. However, this also increases training costs and processing times. In this study, deep sparse learning (DSL) methods are evaluated for the classification of cancer cells and tissues. The Ocelot dataset was used to analyze cell and tissue images. While the highest F1 Score reported in the literature is 75.58%, the proposed method improved the initial F1 Score from 69.45% to 73.12%. Additionally, the DSL model, supported by data augmentation techniques, achieved a 20% improvement in processing time. The findings demonstrate that DSL not only improves accuracy but also reduces processing time, providing more efficient and cost-effective solutions in the field of medical image processing.