Interactive framework for Covid-19 detection and segmentation with feedback facility for dynamically improved...

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Deep Learning

Background:

• Accurate and rapid diagnosis of COVID-19 is critical for pandemic containment.
• Deep learning (DL) and transfer learning (TL) show promise for automated detection and segmentation of infections from medical images.
• Existing AI approaches often utilize Convolutional Neural Networks (CNNs) and Deep Neural Networks (DNNs).

Purpose of the Study:

• To develop a web-based application framework for detecting and segmenting COVID-19 lung infections.
• To implement and evaluate popular DL models, including Mask R-CNN, U-Net, and U-Net++, within the framework.
• To incorporate a feedback mechanism for self-learning and model tuning.

Main Methods:

• A web application framework was designed with a user-friendly interface.
• Variations of Mask R-CNN, U-Net, and U-Net++ models were employed for detection and segmentation.
• Models were trained, evaluated, and tested using Computed Tomography (CT) images from two distinct patient data sources.

Main Results:

• The developed framework successfully processed CT images for COVID-19 detection and segmentation.
• All implemented DL models achieved high performance metrics, including Dice similarity, Jaccard similarity, accuracy, loss, and precision.
• The U-Net model exhibited the highest performance, surpassing 98% accuracy.

Conclusions:

• The proposed web application framework offers an effective tool for COVID-19 lung infection detection and segmentation.
• Deep learning models, particularly U-Net, show significant potential for accurate and efficient analysis of medical images for pandemic diagnosis.
• The framework's self-learning capability can enhance future diagnostic tools for infectious diseases.