Paper Contents
Abstract
Medical image segmentation is a critical component of computer-aided diagnosis and treatment planning, where precise delineation of anatomical structures is essential. This project investigates the performance of the Double U-Net architecturea novel extension of the U-Net model designed to enhance segmentation accuracy for complex medical imaging tasks. Double U-Net leverages two stacked U-Net models along with Atrous Spatial Pyramid Pooling (ASPP) and a pre-trained VGG-19 encoder to capture multi-scale features and improve boundary precision. We compare Double U-Net with traditional segmentation architectures, including U-Net, using four evaluation metrics: Dice Similarity Coefficient (DSC), Mean Intersection over Union (mIoU), Precision, and Recall. Experiments were conducted on three challenging medical datasets spanning various imaging modalities to assess robustness and generalization. Results demonstrate that Double U-Net outperforms conventional models across metrics, particularly on complex datasets with irregular or small target structures, underscoring its potential as a reliable tool for clinical image analysis. This study provides a comprehensive evaluation of Double U-Nets architectural improvements and positions it as a new baseline for medical image segmentation.
Copyright
Copyright © 2024 Manomay Joshi. This is an open access article distributed under the Creative Commons Attribution License.
