Detecting COVID-19 Related Pneumonia On CT Scans Using Hyperdimensional Computing

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Pneumonia is a frequent complication of COVID-19.
• COVID-19 pneumonia exhibits a unique, localized spread pattern in the lungs, differing from typical pneumonia.
• This pattern increases resilience and the risk of acute respiratory distress syndrome.

Purpose of the Study:

• To develop and evaluate a classification algorithm for detecting COVID-19 pneumonia using pulmonary computerized tomography (CT) scans.
• To leverage hyperdimensional computing for accurate and efficient analysis of CT imaging data.
• To explore the potential of identifying uncommon pulmonary diseases for early viral respiratory infection detection.

Main Methods:

• Development of a classification algorithm based on hyperdimensional computing.
• Testing the algorithm on three distinct datasets of pulmonary CT scans.
• Utilizing CT scan features to differentiate COVID-19 pneumonia from other conditions.

Main Results:

• The algorithm demonstrated high performance in detecting COVID-19 pneumonia.
• The highest reported accuracy reached 95.2% with an F1 score of 0.90.
• All tested models achieved a precision of 1, indicating zero false positives.

Conclusions:

• Hyperdimensional computing offers a promising approach for the accurate detection of COVID-19 pneumonia via CT scans.
• The algorithm's high precision suggests its potential for reliable early diagnosis.
• This method could be valuable for identifying novel respiratory viruses through their distinct pneumonia patterns.