Comparative Analysis of Manual and Deep LIIF Extended Global Scoring Method for Ki-67 Across a Spectrum of Oral and...

The Ki-67 labelling index is routinely performed by manual microscopic examination of tissue samples. Digital pathology has improved the efficiency of the diagnostic workflow, however discrepancies between automated and manual estimation of Ki-67 labelling index remain a concern. The present study compared conventional and DeepLIIF-extended global scoring methods for Ki-67 labelling index estimation across a broad spectrum of oral lesions. To our knowledge, this is the first study in oral and maxillofacial pathology to assess the diagnostic precision, reproducibility, and temporal efficiency using DeepLIIF-extended global scoring.

High-power images of Ki-67 stained oral pathologies were analysed using an extended global scoring system by both manual and automated analysis. Manual scoring was performed by two independent pathologists blinded to the clinical diagnosis. Automated scoring was conducted by the DeepLIIF platform. Pearson's and Spearman's correlation coefficients were used to assess the relationship between the automated AI-derived and manual score. Bland-Altman analysis assessed the limits of agreement between the two methods.

Agreement between the AI-derived and manual Ki-67 scores was poor (ICC = 0.389), with a consistent tendency for the AI platform to overestimate Ki-67 positivity (mean bias = - 10.27%). Despite this, AI-derived Ki-67 scores showed a significant positive correlation with manual assessment (Spearman's rho = 0.606, p < 0.0001).

The present study highlights the need for a standardized approach to Ki-67 evaluation in oral and maxillofacial pathology. To create consistent scoring methods, validation protocols, and AI performance objectives for precise assessment of oral lesions, a multidisciplinary agreement is suggested.