Year: 2025
Type: article
Abstract: Tumor edge detection and segmentation are crucial in cancer diagnosis and monitoring.Existing methods are limited to detecting complex tumour edges because of their low spatial resolution.This results in inaccurate detection of tumour boundaries.Deep learning (DL) based approaches have recently emerged as a promising solution for improving the accuracy of tumour detection.We propose an ensemble DL-based U-net and CNN model for tumour edge detection, segmentation, and classification.The model uses Leaky ReLU instead of ReLU and dice loss as a function.It was evaluated on the BraTS 2020 dataset of a diverse range of MR images.It achieves an accuracy of 0.9928, precision of 0.9935, sensitivity of 0.991, and specificity of 0.9978 on BraTS 2020.The algorithm was deployed on a Linux-based embedded Edge AI system combining the processing powers of Nvidia Jetson Nano and Google Coral USB AI Accelerator for faster computation than regular desktops.It is a portable, low-powered, cost-effective, and time-efficient system. .
Source: Allied Medical Research Journal
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7108925344