Anam-Net: Anamorphic Depth Embedding-Based Lightweight CNN for Segmentation of Anomalies in COVID-19 Chest CT Images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Chest computed tomography (CT) is crucial for managing coronavirus disease 2019 (COVID-19).
• Current visual scoring of COVID-19 abnormalities in CT scans is subjective and time-consuming.
• Automated quantification of lung abnormalities in CT images is needed for efficient clinical decision-making.

Purpose of the Study:

• To develop a lightweight, automated method for segmenting COVID-19 related anomalies in chest CT images.
• To introduce Anam-Net, a novel anamorphic depth embedding-based convolutional neural network (CNN).
• To evaluate Anam-Net's performance against state-of-the-art segmentation architectures.

Main Methods:

• Proposed Anam-Net, a lightweight CNN with significantly fewer parameters than existing models like UNet.
• Benchmarked Anam-Net against ENet, LEDNet, UNet++, SegNet, Attention UNet, and DeepLabV3+.
• Deployed Anam-Net on embedded systems (Raspberry Pi 4, NVIDIA Jetson Xavier) and a mobile application (CovSeg) for point-of-care use.

Main Results:

• Anam-Net achieved good Dice similarity scores for segmenting both abnormal and normal lung regions.
• The model demonstrated superior efficiency with 7.8 times fewer parameters than UNet.
• Successful deployment on resource-constrained platforms confirmed its suitability for point-of-care applications.

Conclusions:

• Anam-Net offers a lightweight and effective solution for automated COVID-19 lung abnormality segmentation in CT scans.
• The model's efficiency and portability make it ideal for point-of-care diagnostics.
• The availability of code, models, and a mobile app facilitates further research and clinical adoption.