Authors: Amin Karimi Monsefi, Payam Karisani, Mengxi Zhou, Stacey Choi, Nathan Doble, Heng Ji, Srinivasan Parthasarathy, Rajiv Ramnath

Published on: February 09, 2024

Impact Score: 8.22

Arxiv code: Arxiv:2402.06190

Summary

• What is new: Introduction of LoGoNet with tailored self-supervised learning for medical imaging, leveraging Large Kernel Attention and a dual encoding strategy for better feature extraction without increasing network capacity.
• Why this is important: Challenges in medical applications due to high dataset construction costs and limited labeled data, along with high maintenance costs for daily large data processing.
• What the research proposes: LoGoNet uses a new neural network architecture and a self-supervised learning method that improves medical image segmentation by adeptly capturing long-range and short-range feature dependencies.
• Results: Demonstrated superior performance in inference time and accuracy across standard datasets (BTCV and MSD) compared to eight state-of-the-art models.

Technical Details

Technological frameworks used: Self-supervised learning, multi-task learning framework

Models used: LoGoNet, Vision Transformer (ViT), CNN-based models

Data used: BTCV and MSD datasets

Potential Impact

Medical imaging software markets, healthcare facilities, and companies specializing in medical data processing technologies could benefit or face disruption.

Want to implement this idea in a business?

We have generated a startup concept here: PreciseScan.