For high-speed AI training tool in medical image analysis. The patent covers the techniques deployed by Aiforia's active learning tool, Annotation Assistant, for the accelerated development of AI models for image analysis in pathology.
Remote collaboration with just the click of a button
Our tools for external validation give the end user an easy interface to define validation sets and for collecting validation annotations. The tools also provide a convenient way to invite colleagues or consultants to give blinded scoring or diagnosis according to intended use criteria.
The cloud-based nature of Aiforia Create makes this collaboration even easier. The validation tool then provides a human versus AI comparison and human versus human cross comparison generating important metrics for QA, regulatory or publication documentation.
You can invite validators with a simple click of a button directly from your image
Taking accuracy to new heights
The AI/human versus human cross comparison occurs at the pixel level for semantic segmentation and further using Jaccard index for objects. The pixel level operation is absolutely crucial to robustly test the validity of your AI models. Simple correlation methods have a strong likelihood to overestimate the performance of your AI model and might hide serious errors.
Especially in biology it is important to not only identify the amount of counts but also the exact identity of counts, otherwise the hypothetical situation of for example two different types of equally distributed objects counted from IHC, could yield 100% concordance when validating, even if the actual agreement between operators would be 0%.