Corticosteroid treatment prediction using chest X-ray and clinical data

Area of Science:

• Artificial Intelligence in Medicine
• Medical Informatics
• Computational Biology

Background:

• Long-term complications of COVID-19 are a growing concern, necessitating effective treatments.
• Corticosteroid (CS) treatment is used in the post-acute phase of COVID-19, but its efficacy varies among patients.
• Identifying patients who benefit from CS treatment is crucial to avoid side effects and optimize care.

Purpose of the Study:

• To develop a novel approach for recommending Corticosteroid (CS) treatment for post-acute COVID-19 patients.
• To create a unique dataset combining clinical data (blood tests, spirometry) and X-ray images from 273 patients.
• To leverage advanced machine learning and deep learning models for personalized treatment prediction.

Main Methods:

• Utilized a combination of clinical data and chest X-ray (CXR) images from 273 post-acute COVID-19 patients.
• Developed a novel methodology integrating machine learning and deep learning models, including Vision Transformer (ViT) and InceptionNet.
• Employed advanced preprocessing techniques and pretraining strategies tailored to the specific characteristics of the collected data.

Main Results:

• The integrated approach combining clinical data and CXR images achieved 8% higher accuracy than CXR analysis alone.
• The proposed method demonstrated 80.0% accuracy (78.7% balanced accuracy) and a ROC-AUC of 0.89.
• Identified key factors predictive of long-term complications in post-acute COVID-19 patients.

Conclusions:

• The developed system for CS treatment prediction using neural networks and learning algorithms is unique and effective.
• Demonstrated the efficiency of using mixed data (clinical + imaging) for predicting treatment response in real-world scenarios.
• The system has been deployed in a hospital setting as a recommendation tool, confirming its clinical applicability.