Authors: Fares Bougourzi, Fadi Dornaika, Cosimo Distante, Abdelmalik Taleb-Ahmed

Published on: May 07, 2024

Impact Score: 7.2

Arxiv code: Arxiv:2405.04169

Summary

• What is new: The introduction of the D-TrAttUnet architecture for lesion segmentation, incorporating a composite Transformer-CNN encoder and dual decoders for enhanced precision.
• Why this is important: Lesion segmentation in medical images is highly challenging, requiring advanced machine learning support to assist radiologists.
• What the research proposes: D-TrAttUnet, a novel architecture that includes a hybrid encoder and dual decoders, designed for precise lesion and organ segmentation.
• Results: The proposed model outperformed existing solutions in Covid-19 and Bone Metastasis segmentation tasks and showed adaptability in gland and nuclei segmentation.

Technical Details

Technological frameworks used: D-TrAttUnet architecture with Transformer-CNN encoder and dual decoders

Models used: Encoder-Decoder structure with attention gates

Data used: Data from Covid-19 and Bone Metastasis segmentation tasks, glands and nuclei segmentation tests

Potential Impact

Healthcare and Medical Imaging companies, specifically those specializing in diagnostic imaging and machine learning software for medical applications

Want to implement this idea in a business?

We have generated a startup concept here: MediSeg.