Related Experiment Video
Updated: Oct 12, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
MPS-Net: Multi-Point Supervised Network for CT Image Segmentation of COVID-19
Hong-Yang Pei1,2, Dan Yang1,3, Guo-Ru Liu3,2
1Key Laboratory of Infrared Optoelectric Materials and Micro-Nano DevicesNortheastern University Shenyang 110819 China.
IEEE Access : Practical Innovations, Open Solutions
|November 23, 2021
Summary
A new deep learning network, MPS-Net, effectively segments COVID-19 lung lesions in CT images. This aids in diagnosing and monitoring the pandemic by accurately identifying infection areas.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- The COVID-19 pandemic has caused millions of deaths globally.
- CT imaging reveals distinct lesions characteristic of COVID-19 pneumonia.
- Accurate segmentation of these lesions is crucial for diagnosis and epidemic monitoring.
Purpose of the Study:
- To develop a deep learning model for accurate segmentation of COVID-19 lung lesions in CT images.
- To address challenges posed by diverse lesion shapes and sizes in COVID-19 CT scans.
Main Methods:
- Proposed a novel Multi-Point Supervision Network (MPS-Net) incorporating multi-scale feature extraction and sieve connection structures.
- Implemented a multi-scale input structure to minimize edge loss during convolution.
- Utilized a multi-point supervised training strategy for enhanced segmentation accuracy.
Main Results:
- MPS-Net achieved a Dice Similarity Coefficient (DSC) of 0.8325, sensitivity of 0.8406, specificity of 0.9988, and IOU of 0.742.
- The model demonstrated effective segmentation of COVID-19 infection areas across various lesion sizes.
Conclusions:
- MPS-Net shows significant potential for assisting in the auxiliary diagnosis and treatment of COVID-19.
- The proposed network offers an effective deep learning solution for segmenting COVID-19 lung lesions on CT images.
