Indica
Backed by cutting edge technology, an uncompromising focus on ease-of-use and dedicated customer service, Indica Labs’ software and services are being used to make vital discoveries in pathology labs and research organizations around the world.
+1 (505) 492 0979
sales: info@indicalab.com
support: support@indicalab.com

HALO AI

DEEP LEARNING CLASSIFIER ADD-ON

HALO AI is a collection of train-by-example classification and segmentation tools underpinned by advanced deep learning neural network algorithms. HALO AI classifiers can be trained to quantify tissue classes, to segment tissue classes for analysis with other HALO image analysis modules, to find rare events or cells in tissues, and to categorize cell populations into specific phenotypes. While HALO AI tools handle most research applications, the HALO AI Python plug-in allows you to add additional networks tailored to specific research applications as needed. If you are struggling to achieve accurate tissue classification and cell segmentation with traditional algorithms, HALO AI might be the right solution for you.


Have you been listening to the National Society for Histotechnology HistoHelp podcast series? In their latest podcast, David Krull gives a great intro to AI for Digital Pathology. And even a shout out on the ease of use of HALO AI!

Check out the first episode of HistoHelp Season 2, Artificial Intelligence: Not Scary Afterall to learn how AI is being used to improve patient diagnosis.


HALO AI

DEEP LEARNING CLASSIFIER ADD-ON

HALO AI is a collection of train-by-example classification and segmentation tools underpinned by advanced deep learning neural network algorithms. HALO AI classifiers can be trained to quantify tissue classes, to segment tissue classes for analysis with other HALO image analysis modules, to find rare events or cells in tissues, and to categorize cell populations into specific phenotypes. While HALO AI tools handle most research applications, the HALO AI Python plug-in allows you to add additional networks tailored to specific research applications as needed. If you are struggling to achieve accurate tissue classification and cell segmentation with traditional algorithms, HALO AI might be the right solution for you.


Have you been listening to the National Society for Histotechnology HistoHelp podcast series? In their latest podcast, David Krull gives a great intro to AI for Digital Pathology. And even a shout out on the ease of use of HALO AI!

Check out the first episode of HistoHelp Season 2, Artificial Intelligence: Not Scary Afterall to learn how AI is being used to improve patient diagnosis.


Podcast_Logo

CONTACT  US

For demo, trial, or quotation:
info@indicalab.com
+1(505) 492 0979

Simple & Intuitive Workflow

HALO AI is fully integrated with the intuitive, easy-to-use HALO and HALO Link viewers and employs a simple three-step workflow.  After defining what tissue classes or cell phenotypes you would like to segment, you train the neural network by drawing annotations – no computer programming or AI knowledge required.  Trained classifiers can be applied to segment tissue and cells on any whole slide image or region of interest.

POWERFUL TISSUE SEGMENTATION

HALO AI now includes the option of three powerful neural networks – VGG, DenseNet and MiniNet. VGG, a well known and more traditional network, was used to build the Indica Labs submission in the CAMELYON17 challenge and was the first neural network integrated with HALO AI. DenseNet is a more modern network capable of creating more robust classifiers at higher resolution compared to VGG. MiniNet, a custom network developed at Indica Labs, is more shallow than VGG or DenseNet, but can produce a solution quickly with limited training data and is therefore useful for testing new AI applications. Skip down to see some applications tested with our built-in networks.

EXCEPTIONAL CELL CLASSIFICATION

Segment nuclei with the new Segmentation classifier. Utilize HALO AI’s pretrained networks for H&E, single IHC, or DAPI stained images for an out of the box solution. Or train your own nuclei segmentation network for a specific application (unique tissue or advanced staining protocols). Once nuclei are segmented, take it a step further using the Nuclei Phenotyper classifier to automatically assign cells into user defined phenotypes with a few quick training examples. Skip down to see some applications tested with our Nuclei Segmentor and Nuclei Phenotyper.

custom network integration

AI scientists and developers can make use of our intuitive HALO and HALO Link interfaces to train their own networks using our Python plugin. The plugin generates training patches from HALO AI training annotations and feeds these to your custom script that can link to any neural network framework. Classifiers can then be used in HALO in exactly the same manner as those generated from our own networks – classify whole slide images, use our post-processing tools and select specific tissue class(es) for analysis with any HALO module.

Simple & Intuitive Workflow

HALO AI is fully integrated with the intuitive, easy-to-use HALO and HALO Link viewers and employs a simple three-step workflow.  After defining what tissue classes they would like to segment, pathologists can train the neural network by drawing annotations – no computer programming or AI knowledge required.  Trained classifiers can be applied to segment tissue classes on any whole slide image or region of interest.

POWERFUL TISSUE SEGMENTATION

HALO AI now includes the option of three powerful neural networks – VGG, DenseNet and MiniNet. VGG, a well known and more traditional network, was used to build the Indica Labs submission in the CAMELYON17 challenge and was the first neural network integrated with HALO AI. DenseNet is a more modern network capable of creating more robust classifiers at higher resolution compared to VGG. MiniNet, a custom network developed at Indica Labs, is more shallow than VGG or DenseNet, but can produce a solution quickly with limited training data and is therefore useful for testing new AI applications. Skip down to see some applications tested with our built-in networks.

EXCEPTIONAL CELL CLASSIFICATION

Segment nuclei with the new Segmentation classifier. Utilize HALO AI’s pretrained networks for H&E, single IHC, or DAPI stained images for an out of the box solution. Or train your own nuclei segmentation network for a specific application (unique tissue or advanced staining protocols). Once nuclei are segmented, take it a step further using the Nuclei Phenotyper classifier to automatically assign cells into user defined phenotypes with a few quick training examples. Skip down to see some applications tested with our Nuclei Segmentor and Nuclei Phenotyper.

custom network integration

AI scientists and developers can make use of our intuitive HALO and HALO Link interfaces to train their own networks using our Python plugin. The plugin generates training patches from HALO AI training annotations and feeds these to your custom script that can link to any neural network framework. Classifiers can then be used in HALO in exactly the same manner as those generated from our own networks – classify whole slide images, use our post-processing tools and select specific tissue class(es) for analysis with any HALO module.

HALO AI Applications

Below are just a few examples of the countless applications for HALO AI.

Picture1

VGG

Metastatic Breast Cancer Detection and Staging in H&E-Stained Lymph Nodes

The HALO AI VGG network was employed to detect and stage metastatic tumor cells in H&E-stained lymph nodes of breast cancer patients in this example.  While the VGG network requires more data and training, it works well on examples like this where the stain intensity and morphology are variable.  Download the application note describing the Indica Labs’ submission to the CAMELYON17 project.

islets close-up

VGG

Identification of Islets with Variable Stain and Morphology in H&E-Stained Human Pancreatic Sections

The evaluation and quantification of islets in H&E-stained tissue sections can be very time-consuming due to heterogeneous morphology of islets.  For this application, we train the VGG network to identify islets within human pancreatic tissue sections.  Download the application note describing the method used to build a classifier for accurately segmentation of islets from surrounding exocrine tissue, irrespective of stain or morphological variability.

tumor stroma lymphocyte 400x400

MiniNet

Masking of Tumor, Stroma and Lymphocytic Cell Clusters in H&E-Stained Breast Cancer Sections

For this application, MiniNet was used to separate tumor from lymphocytic clusters and stroma in and H&E stained breast cancer tissue.  MiniNet is a shallow neural network, but it works well on routine classification tasks and may be used as an alternative to HALO tissue classifier (random forest) in some instances.  It is also recommended for quick proof-of-concept studies before moving over to the more data and time intensive VGG and DenseNet networks.  Open full-sized image.

gloms 400x400

DenseNet

Accurate Glomeruli Detection in H&E, Silver, IHC and ISH-Stained Kidney Sections

Here the HALO AI DenseNet neural network was used to segment out glomeruli in differently stained kidney sections, including Jones’ methenamine silver, H&E, ISH and IHC.  In the silver stained slide, the glomeruli and tubules are highly variable in color and morphology.  In the ISH, IHC and H&E-stained sections,  there is little to no color contrast between the cells in the glomeruli and the tubules.  Open full-sized images.

nuclearphenotyper2

Nuclei Phenotyper

Phenotyping of Individual Tumor, Stroma and Lymphocytic Nuclei in H&E-stained Tissue Sections

While the masking networks (MiniNet, DenseNet and VGG) can be trained to segment tissue classes, they cannot be applied to classify individual cells.  For this application, we use the Nuclei Phenotyper.  For this application, nuclei are first segmented using a pre-trained Nuclei Segmentation network and then the individual nuclei are phenotyped using training examples provided by the pathologist (shown here, tumor cells in blue, stroma cells in green and lymphoctyes in yellow). 

Read what our customers are publishing using HALO AI

Our customers are making vital discoveries in oncology, neuroscience, and diabetes research.  Check out some of the research our customers are publishing using the HALO AI Platform.

A Deep Learning Convolutional Neural Network Can Recognize Common Patterns of Injury in Gastric Pathology

Martin, D.R., Hanson, J.A., Gullapalli, R.R., Schultz, F.A., Sethi, A. and Clark, D.P. Archives of Pathology & Laboratory Medicine In-Press | Published April 5, 2019 | DOI: 10.5858/arpa.2019-0004-OA Context.— Most deep learning (DL) studies have focused on neoplastic pathology, with...

Detection of Lung Cancer Lymph Node Metastases from Whole-Slide Histopathologic Images Using a Two-Step Deep Learning Approach

Hoa Hoang Ngoc Pham, Mitsuru Futakuchi, Andrey Bychkov, Tomoi Furukawa, Kishio Kuroda, Junya Fukuoka The American Journal of Pathology | Volume 189, Issue 12, December 2019 | DOI: 10.5858/arpa.2019-0004-OA The application of deep learning for the detection of lymph node metastases...

Prognostic significance of mesothelin expression in colorectal cancer disclosed by area-specific four-point tissue microarrays

Takehiro Shiraishi, Eiji Shinto, Ines P. Nearchou, Hitoshi Tsuda, Yoshiki Kajiwara, Takahiro Einama, Peter D. Caie, Yoji Kishi & Hideki Ueno Virchows Archiv | February 2020 | DOI: 10.1007/s00428-020-02775-y Mesothelin (MSLN) is a cell surface glycoprotein present in many cancer...

Immunogradient Indicators for Antitumor Response Assessment by Automated Tumor-Stroma Interface Zone Detection

Allan Rasmusson, Dovile Zilenaite, Ausrine Nestarenkaite, Renaldas Augulis, Aida Laurinaviciene, Valerijus Ostapenko, Tomas Poskus, and Arvydas Laurinavicius The American Journal of Pathology, Vol. 190, No. 6, June 2020 | DOI: doi.org/10.1016/j.ajpath.2020.01.018 The distribution of tumor-infiltrating lymphocytes (TILs) within the tumor...

Platform Compatibility

HALO AI is compatible with all of the file formats that can be used in HALO and HALO Link.  Not on the list? Email us your requirements.

File Formats

Non-proprietary (JPG, TIF)
Nikon (ND2)
3D Histech (MRXS)
Perkin Elmer (QPTIFF, component TIFF)
Olympus (VSI)
Hamamatsu (NDPI, NDPIS)
Aperio (SVS, AFI)
Leica (SCN, LIF)
Ventana (BIF)

HALO AI Collaboration

Want to work with external pathologists or translational research groups on a HALO AI project? No problem. Our browser-based image management system, HALO Link, is fully integrated with HALO AI to facilitate this collaboration. Once you’ve invited your collaborator to a HALO AI project, they can upload their own images and draw training annotations, nothing required except internet access and a browser. 

Request a demonstration of HALO AI CLOUD, our own web-hosted HALO AI and HALO Link deployment, using the form below or send your request by email to info@indicalab.com

HALO AI Collaboration

Want to work with external pathologists or translational research groups on a HALO AI project? No problem. Our browser-based image management system, HALO Link, is fully integrated with HALO AI to facilitate this collaboration. Once you’ve invited your collaborator to a HALO AI project, they can upload their own images and draw training annotations, nothing required except internet access and a browser. 

Request a demonstration of HALO AI CLOUD, our own web-hosted HALO AI and HALO Link deployment, using the form below or send your request by email to info@indicalab.com

Want to see more?

Fill out the form below to request a recorded or live demo of HALO AI or any of our other products or services.
You can also drop us an email at info@indicalab.com

Your Name (required)

Your Email (required)

Your State/Country

Which of the following products are of interest?


Tell us a bit more about your request.

RELATED RESOURCES

Webinar Recording

RELATED PRODUCTS

HALO image analysis
HALO analysis modules
HALO Link image mangement
PHARMA SERVICES

CONTACT  US

For demo, trial, or quotation:
info@indicalab.com
+1(505) 492 0979