Radiomics-Based Detection of COVID-19 from Chest X-ray Using Interpretable Soft Label-Driven TSK Fuzzy Classifier

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Chest X-Ray (CXR) is a widely accessible imaging modality for diagnosing respiratory conditions.
• Existing AI models for COVID-19 detection from CXR lack interpretability due to complex deep features.

Purpose of the Study:

• To develop an interpretable AI model for COVID-19 detection using CXR radiomics.
• To improve classification accuracy in multi-categorical tasks (COVID-19, No-Findings, Pneumonia).
• To address the interpretability limitations of current deep learning models.

Main Methods:

• Development of a Takagi-Sugeno-Kang (TSK) fuzzy system for COVID-19 detection.
• Extraction of radiomics features from Chest X-Ray (CXR) images.
• Application of a soft label matrix transformation and a compactness class graph to enhance classification and prevent overfitting.

Main Results:

• The proposed TSK fuzzy system achieved over 83% classification accuracy on a dataset of 600 CXR images.
• The model demonstrated superior performance compared to five state-of-the-art AI models.
• The system maintained high interpretability, a significant advantage over black-box models.

Conclusions:

• The interpretable TSK fuzzy system offers a promising approach for accurate and understandable COVID-19 detection from CXR.
• The novel methods of soft label transformation and compactness class graph effectively improve classification accuracy and model robustness.
• This interpretable AI model can aid in early isolation of suspected COVID-19 cases, improving public health response.