Predicting Hospitalization Length in Geriatric Patients Using Artificial Intelligence and Radiomics

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Accurate prediction of COVID-19 patient hospitalization duration is vital for resource allocation.
• Radiomics, extracting quantitative features from CT scans, combined with machine learning (ML) presents a promising predictive approach.

Purpose of the Study:

• To adapt and utilize ML architectures with CT radiomics data.
• To analyze algorithm capabilities in predicting hospitalization length at patient admission.

Main Methods:

• 168 COVID-19 patients' CT lung images were segmented to isolate ground glass areas.
• 92 radiomics features were extracted after filtering and resampling, followed by LASSO for feature reduction.
• Three ML classifiers (LSVM, MNN, ESD) were trained and validated using 5-fold cross-validation.

Main Results:

• The linear support vector machine (LSVM) achieved the highest accuracy (86.0%) and AUC (0.93).
• Medium neural network (MNN) and ensemble subspace discriminant (ESD) also demonstrated reliable predictive performance.

Conclusions:

• Radiomic features can form the basis of ML frameworks for predicting COVID-19 hospitalization duration.
• Radiomics-based ML models show potential for accurate prediction of patient hospitalization length.