Determinants and limitations of subjective and partially objective methods when analysing Meibomian gland loss

To compare the advantages and limitations of subjective and partially objective methods for assessing Meibomian gland loss (MG loss) and to identify possible associated factors.

MG loss was assessed subjectively using meibographic images of the eyelids by six experienced clinicians. Additionally, using three partially objective methods: automatic MG loss detection (M1), segmentation of MG loss using the meibography system (M2) and with independent image analysis software (M3). MG loss was classified into five levels using the Pult Meiboscale. The mean subjective grade ( $\overline{\mathrm{ECG}}$ ) for each image was computed from the individual subjective grade (experienced clinician grading, ECG) and assumed to be the ground truth. The eyelid eversion quality (EEQ) was classified into three grades.

A total of 28 subjects were analysed (mean age: 26.7 ± 7.8 years; 17 females and 11 males). Among the partially objective methods, only M3 demonstrated significant reliability with $\overline{\mathrm{ECG}}$ for the upper (k = 0.55, p = <0.001) and lower eyelids (k = 0.23, p = 0.047). Overall, the ECG demonstrated better reliability with $\overline{\mathrm{ECG}}$ compared to the partially objective methods. The partially objective methods tended to overestimate MG loss for lower values while underestimating them for higher values. Significant differences in MG loss were found between males and females (p = 0.001).

Partially objective assessment of MG loss did not surpass subjective evaluation by an experienced clinician, with the EEQ being identified as a crucial factor in the assessment of the MG loss. This factor should be considered for improving evaluation accuracy. These findings underscore the assessment of MG loss, considering the proficiency and experience of the clinician, as well as the urgent need to improve standardisation, training and advancements in objective methods to refine the evaluation of MG loss.