Medical AI platform HALO AI

for medical researchfor histopathology laboratoriesmedical imaging

Overview
HALO AI from Indica Labs is a trainable artificial intelligence platform for digital pathology enabling segmentation, classification, and phenotyping. Underpinned by modern deep learning networks, it is tuned for brightfield and fluorescence applications via a train-by-example interface. The intuitive workflows require no programming or AI knowledge and allow users to define tissue classes and cell phenotypes or to train neural networks by drawing annotations.

Integration
HALO AI is integrated with HALO Link and HALO platforms to support collaborative image management and quantitative image analysis workflows.

Key features

• Train-by-example interface for easy creation of custom classifiers and segmentors.
• AI-powered annotation tool to rapidly acquire training annotations by pointing and clicking objects of interest.
• Real-time tuning that shows network training progress live, allows interactive parameter changes, and supports selection of probability cut-offs using heatmap outputs.
• Supports probability map outputs as an alternative to traditional masks for performance evaluation.
• Interactive markups enabling toggling of populations of interest, combinable with probability thresholding for validation exploration.
• Pre-trained nuclear and membrane segmentors available; HALO AI enables optimization for bespoke applications when needed.
• Robust to variability in morphology, staining protocols, tissue quality, and uneven staining; can be trained across diverse stains (examples: PAMS, Trichrome, H&E, IHC).
• Ability to chain multiple HALO AI classifiers into classifier pipelines for complex workflows.

Apps (pre-trained HALO AI classifiers / phenotypers)

• Breast IHC Tumor Tissue Detection — pre-trained classifier to detect, segment, and quantify tumor and other areas in hematoxylin and DAB-stained whole-slide breast cancer images.
• NSCLC IHC Tumor Tissue Detection — pre-trained classifier to detect, segment, and quantify tumor and non-tumor areas in hematoxylin and DAB-stained NSCLC whole-slide images.
• NSCLC IHC Cancer Cell Phenotyper — pre-trained object phenotyper to detect, segment, and quantify non-cancer cells, IHC-positive cancer cells, and IHC-negative cancer cells in NSCLC.
• Pan Cancer H&E Lymphocyte Cell Phenotyper — pre-trained object phenotyper to detect and quantify lymphocytes across whole-slide H&E images of multiple tumor types.
• Gastric H&E Tumor Tissue Detection — masking classifier designed to segment tumor, stroma, necrosis/other, and glass areas in gastric cancer H&E images.
• HNSCC H&E Tumor Tissue Detection — masking classifier for head & neck squamous cell carcinoma H&E images to segment tumor, stroma, necrosis/other, and glass areas.
• NSCLC H&E Tumor Tissue Detection — masking classifier for NSCLC H&E whole-slide images.
• Ovarian H&E Tumor Tissue Detection — masking classifier for ovarian cancer H&E whole-slide images.

Usage and outputs
HALO AI networks, once trained, can be incorporated into HALO modules to maximize utility across analysis pipelines. Outputs include segmentation masks and probability maps; probability thresholding and interactive markups can be used to refine and validate results.

Regulatory note
For Research Use Only. Not for Use in Diagnostic Procedures.

Technical specifications

• Train-by-example deep learning for brightfield and fluorescence images.
• AI annotation tool for rapid ground-truth generation.
• Real-time training visualization and on-the-fly parameter tuning.
• Supports probability map outputs and traditional mask outputs.
• Pre-trained nuclear and membrane segmentation models available; custom training supported for bespoke use cases.
• Designed to handle variable staining types and tissue quality (examples: PAMS, Trichrome, H&E, IHC).
• Integrates with HALO Link for collaborative image management and study-based training data collection.