(P90) The Impact of Artificial Intelligence Assistance on Interobserver Agreement in ER, PR, HER2, and Ki-67 Assessment
Fadime Gul Salman (Turkiye) et al.
In this study, Salman et al. evaluated how the Aiforia® Breast Suite affects agreement among pathologists scoring breast cancer biomarkers in 296 cases. The use of AI increased diagnostic consistency and reduced scoring discrepancies for ER, PR, and HER2, notably decreasing complete disagreement for HER2 from 21% to 14%.
(P147) CAD-assisted histological diagnosis in prostate mapping: a multicentric Italian experience of the AI-FLOPP study
Moira Ragazzi (Italy) et al.
(Abstract presentation) A Novel Prediction tool (AIFORIA) for Classifying Spindle Cell Lesion subtypes using Deep Learning on H&E Sections
Kristijan Skok (Austria) et al.
June 19 at 14:30 CEST
During the “Machine Learning and Algorithm Development II" session
In this proof-of-concept study, Skok et al. evaluated (or compared?) the performance of two AI models developed with Aiforia® Create, using standard CNNs and Aiforia's proprietary Foundation Engine, to classify complex spindle cell lesion subtypes from routine H&E slides via a weakly supervised approach. The standard CNN-based AI model showed strong sensitivity for solitary fibrous tumors (SFT) but struggled with class imbalance. The AI model built on Foundation Engine provided a more balanced predictive approach, achieving an overall concordance of 64.9% while maintaining solid detection rates for Synovial Sarcoma (66.2% recall). However, accurately identifying MPNST cases remains a complex challenge across all tested architectures, highlighting the need for further dataset expansion and targeted model optimization to enhance clinical applicability.