Radiomics-based fundus autofluorescence analysis in central serous chorioretinopathy-MICRoN report number twelve

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Central serous chorioretinopathy (CSCR) is a condition affecting vision, with varying subtypes.
• Accurate classification of CSCR subtypes is crucial for effective patient management and treatment strategies.
• Current methods for CSCR subtyping may lack quantitative precision.

Purpose of the Study:

• To apply radiomics feature extraction to fundus autofluorescence (FAF) images for automated CSCR subtype identification.
• To identify quantitative imaging biomarkers capable of distinguishing simple versus complex and acute versus chronic CSCR.
• To evaluate the performance of machine learning classifiers in predicting CSCR subtypes based on radiomic features.

Main Methods:

• Radiomic features were extracted from 96 FAF images of CSCR patients using Pyfeats.
• Feature selection was performed by excluding zero-variance and highly correlated variables (Pearson's r > 0.8).
• Logistic regression, random forest, and extreme gradient boosting (XGBoost) classifiers were trained and evaluated for CSCR stage prediction.

Main Results:

• For simple versus complex CSCR, eight selected features achieved a mean AUC of 0.90 and accuracy of 0.80 with logistic regression.
• For acute versus chronic CSCR, ten features were selected, and the XGBoost model achieved a mean AUC of 0.69 and accuracy of 0.71.
• The radiomic approach demonstrated robust classification performance, particularly for distinguishing simple and complex CSCR.

Conclusions:

• Radiomic features extracted from FAF images serve as effective quantitative biomarkers for automated CSCR classification.
• This methodology supports the potential for automated disease staging and subtyping in central serous chorioretinopathy.
• The findings highlight the utility of radiomics in enhancing diagnostic precision for CSCR.