MT-nCov-Net: A Multitask Deep-Learning Framework for Efficient Diagnosis of COVID-19 Using Tomography Scans.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Accurate segmentation of COVID-19 lesions in CT scans is crucial for computer-aided diagnosis.
• Deep learning (DL) shows promise but faces challenges like data heterogeneity, lesion variability, and limited annotations.

Purpose of the Study:

• To propose a novel multitask regression network, MT-nCov-Net, for efficient COVID-19 lesion segmentation.
• To address the limitations of existing DL approaches in COVID-19 diagnosis.

Main Methods:

• Developed MT-nCov-Net, a multitask shape regression network for lesion segmentation.
• Incorporated a multiscale feature learning (MFL) module to capture semantic information across scales.
• Introduced a fine-grained lesion localization (FLL) module with an adaptive dual-attention mechanism.

Main Results:

• MT-nCov-Net effectively segments COVID-19 lesions by regressing their shape, learning complete lesion properties.
• The framework demonstrated superior performance over current state-of-the-art methods on two public datasets.
• Validated effectiveness in tackling challenges in COVID-19 diagnosis.

Conclusions:

• MT-nCov-Net offers an effective solution for COVID-19 lesion segmentation in CT scans.
• The proposed approach enhances computer-aided diagnosis systems for infectious diseases.
• Multitask learning and attention mechanisms improve segmentation accuracy and efficiency.