Reinforcement Learning Based Diagnosis and Prediction for COVID-19 by Optimizing a Mixed Cost Function From CT Images

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Infectious Disease Diagnostics

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Existing diagnostic methods face challenges in speed and efficiency.
• Timely detection is crucial for controlling the spread of novel coronavirus disease.

Purpose of the Study:

• To develop a reinforcement learning-based framework for efficient COVID-19 detection.
• To create a prediction framework for patient disease progression using parameter sharing.
• To enhance COVID-19 diagnosis and prognosis through advanced machine learning techniques.

Main Methods:

• A novel detection framework utilizing reinforcement learning and a mixed loss function.
• Parameter sharing across multiple detection frameworks for disease progression prediction.
• Development of a high-quality CT image dataset for training and validation.

Main Results:

• Achieved 98.31% classification accuracy for COVID-19 detection.
• Demonstrated high performance in precision (98.82%), sensitivity (97.99%), specificity (98.67%), and AUC (0.989).
• External validation accuracy reached 93.34% and 91.05%; prediction framework accuracy was 91.54%.

Conclusions:

• The proposed reinforcement learning framework is effective and robust for COVID-19 detection.
• The integrated prediction framework accurately forecasts disease progression without additional training.
• This AI-driven approach offers a significant advancement in managing the COVID-19 pandemic.