Login Search
FMDB Transactions on Sustainable Health Science Letters Submit Paper
    Crossref logo

    Archiving

    Portico

    Fernando Martins De Bulhão is a member of

    Crossref logo

Domain-Adaptive Multi-Stage Hybrid Segmentation with Deep Learning and Watershed Techniques for Accurate Medical Image Analysis

Authors:
Vishal Kumar Kanaujia, Awadhesh Kumar, Satya Prakash Yadav

Addresses:
Department of Computer Science and Engineering-Data Science, ABES Engineering College, Ghaziabad, Uttar Pradesh, India. Department of Computer Science and Engineering, Kamla Nehru Institute of Technology, Sultanpur, Uttar Pradesh, India. Department of Computer Science and Engineering, Madan Mohan Malaviya University of Technology, Gorakhpur, Uttar Pradesh, India.

Abstract:

This paper introduces a unique framework for kidney tumour segmentation that leverages deep learning and traditional image processing methods, focusing on the KiTS19 dataset. The proposed approach combines the U-Net architecture with the watershed and random walker algorithms to enhance boundary refinement and segmentation accuracy. To address domain adaptation issues and ensure reliable performance across datasets, a Gradient Reversal Layer (GRL) is also included. The hybrid segmentation model achieves 94.35% accuracy, 93.10% F1-score, and 89.50% Intersection over Union (IoU) on the source domain, demonstrating better performance than conventional deep learning architectures. On unseen target data, the model retains a high accuracy of 91.25%, highlighting the efficacy of domain adaptation strategies. The suggested model has better generalisation and boundary-refinement capabilities than current methods such as U-Net and Fully Convolutional Networks (FCN), making it a good fit for clinical applications that demand accurate segmentation. This multi-stage hybrid technique bridges the gap between deep learning and traditional image processing by improving generalisation performance and ensuring improved border delineation. The findings validate the importance of combining deep learning with domain-adaptation strategies to achieve accurate, generalizable segmentation solutions.

Keywords: Kidney Segmentation; Image Processing; Kidney Tumour Segmentation; Fully Convolutional Networks (FCN); Kidney Cancer; Refinement and Generalisation.

Received on: 02/03/2025, Revised on: 05/05/2025, Accepted on: 23/07/2025, Published on: 08/03/2026

DOI: 10.69888/FTSHSL.2026.000592

FMDB Transactions on Sustainable Health Science Letters, 2026 Vol. 4 No. 1, Pages: 12-30

  • Views : 55
  • Downloads : 8
Download PDF