XAI techniques with clinical markers found to predict COVID-19 severity

In a recent study titled Explainable artificial intelligence approaches for COVID-19 prognosis prediction using clinical markers, researchers demonstrate that employing explainable AI (XAI) techniques with clinical and laboratory markers could enhance the prediction of COVID-19 severity in healthcare facilities.

These models offer valuable insights into clinical markers, enabling healthcare professionals to optimise the allocation of critical medical resources such as ICU beds, ventilators, and medications. Additionally, the models can serve as a supportive tool for obtaining a second opinion.

The research used multiple machine learning and deep learning models to forecast severe COVID-19 cases in advance, while data sets were obtained from two Indian hospitals: Dr TMA Pai Hospital and Kasturba Medical College. The dataset comprised 599 non-severe patients and 300 severe patients requiring ICU admission.

After training and testing the machine learning and deep learning models, five XAI techniques—SHAP, LIME, Eli5, QLattice, and Anchor—were employed for interpretation. The XAI approaches revealed the most important markers in predicting a patient’s severity to be c-reactive protein (CRP), lymphocytes, basophils, albumin, D-Dimer, NLR, and neutrophils.

Looking ahead, the authors recommend the implementation of cloud-based models for more efficient storage of data and code. They propose the use of high-performance GPUs for training deep learning algorithms and suggest aggregating electronic health records from various hospitals across different countries before model training.