Document Type

Article

Publication Date

3-2026

Abstract

Accurate segmentation of histopathological images is essential to support early diagnosis and effective treatment planning in cancer care. This study presents CancerSeg-XA, a deep learning-based histopathology segmentation system designed to deliver robust performance across diverse tissue types and imaging sources. Built upon the DeepLabV3+ framework, CancerSeg-XA incorporates architectural enhancements to strengthen feature representation and improve model stability. The system was evaluated on three widely recognized datasets—BCSS, PanNuke, and PUMA—each presenting distinct structural and clinical challenges. Across all datasets, CancerSeg-XA consistently outperformed the baseline DeepLabV3+ in terms of segmentation accuracy, recall, and F1-score. Specifically, it achieved accuracy improvements of 4.78%, 4.31%, and 3.22% on BCSS, PanNuke, and PUMA, respectively, along with substantial gains in FwIoU. These results highlight the model’s ability to generalize effectively across varied histopathological contexts, positioning CancerSeg-XA as a promising solution for clinical integration and future research in digital pathology.

Download

Share

COinS