ML-guided analysis of PO2-regulated RBC capillary velocity for ME/CFS classification. (A) Flow diagram of ML-guided analysis. Various physical parameters or features from PO2-regulated RBC capillary velocity were used as input for different ML algorithms to calculate ME/CFS correlation and diagnostic accuracy. Slope was calculated based on RBC velocities at 4 PO2 levels, whereas slope-range was calculated based on RBC velocity within a range from 25% to 75% of the maximum velocity. Subslopes 1 to 3 were calculated based on velocities of 2 adjacent PO2 levels. R2 was the coefficient of determination calculated based on RBC velocities at 4 PO2 levels. Velocities 1 to 4 represented RBC velocities at PO2 levels of 34, 25, 12, and 0 mm Hg, respectively. Slope-old represented the slope calculated based on the velocities of RBCs falling within the lowest 50% of the velocity range. Maximum subslope was the maximum value of all the subslopes. (B) Correlation heat map between features and ME/CFS. Slope showed the highest correlation (0.548) for ME/CFS classification. (C) Classification accuracy using different subsets of the features and ML algorithms. Combination of features 2 to 6 showed the highest accuracy in KNN (0.778). Calculation was conducted based on data collected from 23 healthy participants and 35 patients with ME/CFS.
