Novel COVID-19 Diagnosis Delivery App Using Computed Tomography Images Analyzed with Saliency-Preprocessing and Deep...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Pandemic Response Technologies

Background:

• Remote delivery of diagnoses and disease progression information is crucial for managing COVID-19.
• Chest computed tomography (CT) scans are vital for assessing COVID-19 pneumonia.
• Minimizing risk during pandemics requires efficient data sharing and analysis.

Purpose of the Study:

• To develop an application for remote delivery of diagnoses and disease progression information for COVID-19 patients.
• To focus on image preprocessing techniques for identifying and highlighting ground glass opacity (GGO) and pulmonary infiltrates (PIs) in CT scans.
• To utilize Convolutional Neural Networks (CNNs) for classifying pneumonia disease progression in COVID-19 cases.

Main Methods:

• Image preprocessing techniques were applied to CT scan sequences of COVID-19 cases.
• Saliency map fusion was used to highlight GGO and PI patterns.
• CNNs were trained and tested on a three-class classification scheme for disease progression.
• Data was shared via an application among patients, respiratory triage/radiologists, and multidisciplinary teams.

Main Results:

• The CNN classification scheme achieved a macro-precision of over 94.89% in a two-fold cross-validation for three-class, disease-level COVID-19 classification.
• Segmentation and classification results demonstrated comparability to those of a medical specialist.
• The application facilitated understanding of disease severity through CT analysis and medical diagnosis.

Conclusions:

• The developed AI-powered application effectively analyzes COVID-19 CT scans to identify and classify disease progression.
• The system provides accurate, specialist-comparable results, enhancing remote diagnostic capabilities.
• This technology aids in managing COVID-19 and can be applied to future pandemics.