HALO AI | Train-by-Example Analysis Software

Collaborate on the development of AI using the HALO Link collaborative image management platform. Simply invite your colleagues to a study and have them add training data.

High levels of variability are often encountered in image analysis but is no match for HALO AI. Common sources of variability include diverse morphologies, alterations of morphologies from staining protocols, differences in tissue quality, uneven staining, and more.

HALO AI can be easily trained to accommodate variability to deliver accurate segmentation and classification results across large studies. HALO AI can even be trained to work across vastly different stains such as PAMS, Trichrome, H&E, and IHC.

Pre-trained nuclear and membrane segmentors available in HALO are advanced tools for nuclear and membrane segmentation, but when you need to optimize a network for a bespoke application, you need HALO AI.

As with the tissue classification and segmentation networks, you can quickly add training data with the AI-based annotation tool and train the network to optimally segment nuclei or cell membranes. And once trained, any HALO AI network can be incorporated into HALO modules for maximum utility.

Quickly acquire training annotations with the new AI annotation tool in HALO AI. Simply point and click your object of interest to add an annotation. Create classifier pipelines by connecting multiple HALO AI classifiers into a single workflow.

Use real-time tuning in HALO AI to watch a network as it trains in real time. Toggle the mark up on and off to evaluate performance, choose to add training data, or change parameters on-the-fly.

Once a HALO AI model is trained, a probability map can be used as an alternative output to a traditional mask to evaluate performance. Use real-time tuning to select an appropriate probability cut-off for a given class and view the output in a heatmap.

Investigate results with interactive markup images where you can toggle on and off each population of interest. Interactive markups can be combined with probability thresholding and are especially valuable in exploring validation outputs.