Mindpeak

Bringing together technology and healthcare, to deliver better outcomes

Mindpeak is now our esteemed partner where Crosscope will integrate MINDPEAK’S BREAST SUITE contains algorithms for Ki-67, ER, PR, and HER2, intended to assist pathologists in the primary diagnosis of breast cancer cases. With the deep workflow integration into the Crosscope Dx platform, the correct algorithms are automatically chosen and the whole slide images pre-analyzed in the background directly after scanning with the interplay of several AI algorithms:

Tissue is found and tumorous areas are identified directly on the IHC image
Depending on the biomarker, either hotspots for scoring are proposed (Ki-67) or a score for the whole tumor is given (ER, PR, HER2) – always with the clinical score. The pathologist stays in full control and can add/remove areas/cells, if necessary.
Within these tumor areas, the tumor cells are found and classified into positive (red) and negative (yellow) displayed with a colored dot per nucleus. All non-tumor cells (e.g. stroma) are ignored

Within these tumor areas, the tumor cells are found and classified into positive (red) and negative (yellow) displayed with a colored dot per nucleus. All non-tumor cells (e.g. stroma) are ignored

pre-calculation in the background without perceivable time delay
fully automated (0-click)
works on single-cell level
tumor/stroma differentiation
works out of the box with real-life image variability

Performance: At least 60 % faster with Mindpeak’s algorithms

Mindpeak’s Breast Suite passed a German proficiency test, intended for laboratories to renew their accreditation, with 96 % concordance.

In addition, the algorithms were validated in the largest variability study to date for a deep learning AI in digital pathology showing non-inferiority between human alone vs. “human + AI” with 204 cases stained on three stainers (Dako, Leica, Roche) and digitized on six instruments (five scanners: 3DHistech, Hamamatsu, Leica, Philips, Roche, and one microscope camera: Nikon).

This study, currently under publication in a peer-reviewed journal, proves the robustness of the algorithms against real-live image variability and that the AI algorithms work out-of-the-box in any laboratory setting without retraining.