Abstract
BACKGROUND: Ki67 is a prognostic factor in breast carcinoma (BC). We evaluate algorithm-based digital image analysis for Ki67 proliferation index (KPI). METHODS: A retrospective study was conducted on 81 BC cases. KPI (by eyeballing, EB) and tubule scores were obtained from records. A region of interest within the hotspot on the scanned slide was annotated and analysed using ImageJ (Reference standard) and Optrascan algorithm (OA). Statistical analysis included Pearson correlation (PC) included Pearson Correlation (PC) and Bland - Altman plots (BAP). RESULTS: EB and OA correlated with ImageJ (p < 0.001). OA had a stronger correlation (r: 0.9913) than EB. BAP revealed larger bias with EB (bias: 1.309) than OA. EB showed significant bias in tumours with >10% tubule formation (bias: 7.268) compared with OA. CONCLUSION: OA is a reasonable alternative for KPI. EB tends to overestimate in solid tumours and underestimate in differentiated tumours. OA overestimates with increased nuclear pleomorphism and underestimates in increased intra-tumoural lymphocytes or fibroblasts.