AIM-HER2 Breast Cancer
AI-based Measurement of HER2 Scoring
AIM-HER2 delivers automated and reproducible digital HER2 scoring with explainable AI heatmap visualizations. Now available on the AISight™ Image Management System.1
Confidently score HER2
Accurate and reproducible HER2 scoring to assist pathologists, especially in reviewing borderline cases and HER2 low expression.
Streamline your workflow
Multi-scanner compatibility plus automated whole slide image analysis and control tissue identification — no manual ROI selection.
Interpret and Explain
Additive MIL based heatmaps2 highlight features that drive HER2 score prediction, enabling human pathologist explainability.
Specifications
- Intended Use: Research Use Only
- Indications: Breast Cancer
- Clones: Ventana 4B5 and Dako HercepTest™
- Scanners: Leica Aperio® AT2 and GT450; Hamamatsu NanoZoomer® s360
- Inputs: Breast cancer biopsy, resection, and/or excision sample from primary, recurrent, or metastatic tumor (excluding in situ tumor)
- Outputs: HER2 score (0, 1+, 2+, 3+); Area of invasive carcinoma; Additive multiple instance learning (aMIL) density heatmap
AIM-HER2 Breast Cancer may be compatible with additional scanner types. Please contact us for more information.
AIM-HER2 Scoring Heatmaps
Additive MIL based heatmaps2 highlight tissue patterns features that most contribute to AIM-HER2's score prediction. Pathologists are able to more confidently assess HER2 expression, especially in borderline cases with more slide-level heterogeneity as they can quickly and more easily focus on the most critical features and slide areas.
“I like that these heatmaps correlate to the likelihood that the region is a particular score. It gives me the immediate intuition as to the parts of the slide I should focus on.”
— Health System Pathologist, UK Cancer Center
Rigorously trained and tested
>157K
Annotations
12K
Slide-level HER2 scores
4K
Slides
65+
Board-certified breast pathologists
Resources
1AIM-HER2 is for Research Use Only. Not for use in diagnostic procedures.
2 Javed, et al. Additive MIL: Intrinsically Interpretable Multiple Instance Learning from Pathology. Computer Vision and Pattern Recognition, June 2022. arXiv:2206.01794