Project Spotlight: SpIntellx

by Ceren Tusmen, PhD

SpIntellx is a computational and systems pathology company providing explainable AI guides to pathologists and biopharma companies. The project, initially named SPDx, is a spinout from the Center for Commercial Applications of Healthcare Data. Led by Michael J. Becich, MD, PhD; S. Chakra Chennubhotla, PhD; Jeffrey L. Fine, MD; D. Lansing Taylor, PhD;, and A. Burak Tosun, PhD, the team is developing novel tools, HistoMapr™ and TumorMapr™, with the aim of providing personalized spatial intelligence to pathologists and biopharma companies.

SpIntellx is a computational and systems pathology company providing explainable AI guides to pathologists and biopharma companies. The project, initially named SPDx, is a spinout from the Center for Commercial Applications of Healthcare Data. Led by Michael J. Becich, MD, PhD; S. Chakra Chennubhotla, PhD; Jeffrey L. Fine, MD; D. Lansing Taylor, PhD;, and A. Burak Tosun, PhD, the team is developing novel tools, HistoMapr™ and TumorMapr™, with the aim of providing personalized spatial intelligence to pathologists and biopharma companies.

Each year, 1.6 million breast biopsies are performed in the US to diagnose cancer as a follow-up to suspicious findings through diagnostic tests, such as mammograms. During a breast biopsy, a tissue specimen from the concerning area is removed, chemically and physically processed, fixed on a slide, and stained to be examined under a microscope by a pathologist. These histological samples collected from patients fall into a wide spectrum, ranging from benign to high-risk lesions to cancer, and this characterization process can be highly subjective depending on the pathologist making the decision. According to a recent study supported by the National Cancer Institute and the National Cancer Institute-funded Breast Cancer Surveillance Consortium, there is substantial discordance between pathologists’ diagnoses – particularly for the high-risk lesions. There is need for more consistent and reliable methods of characterization to avoid chances of misdiagnosis, which can result in over or under-treating the patient.

 

Left to right, A. Burak Tosun, PhD, S. Chakra Chennubhotla, PhD, and D. Lansing Taylor, PhD, testing the xAI feature of HistoMaprTM on a breast tissue sample.

There have been efforts to develop pathology platforms that digitize pathology images taken through a microscope that would be traditionally viewed by a pathologist. These same platforms can use digital analytics tools or advanced computational pathology techniques that apply artificial intelligence (AI) and machine learning to these digitized data. The goal is to improve the process of decision-making in pathology to deliver the best care to patients. The challenge, however, is that tumor growth is a complex and dynamic process with many biological and chemical constituents, including different cell types, vascular networks consisting of the lymphatic and blood vessels, and the components of the extracellular matrix. The interaction among these cellular and non-cellular structures within the tumor niche is defined as the tumor microenvironment (TME), which is now known to play a critical role in the development of malignancies. When the TME is in a healthy state, it protects against cancer, and when it’s not, it can support cancer. Existing methods of digital pathology are limited in the ability to characterize these critical changes in the TME with a spatial context; in other words, how the cells and the biological components are oriented to each other. With greater spatial knowledge of the tumor microenvironment, there will be a higher level of breast cancer prognostic and diagnostic capability.

A team of clinicians and scientists at the University of Pittsburgh School of Medicine, consisting of S. Chakra Chennubhotla, PhD; D. Lansing Taylor, PhD; Michael J. Becich, MD, PhD; Jeffrey L. Fine, MD; and A. Burak Tosun, PhD, has developed tools called HistoMaprTM and TumorMaprTM that successfully address and overcome these issues. SpIntellx’s technology is unique in two main ways:

  1. SpIntellx builds computational models based on spatial intelligence, or the understanding of spatial relationships among the cellular and non-cellular structures in a TME, going beyond methods that just identify structures and count cells in a digitized image of a pathology sample.
  2. SpIntellx applies explainable AI (xAI) to digitized pathology images. xAI is an emerging concept referring to “AI that can justify its results with data” to help the pathologist understand the mechanism by which they make a decision towards a diagnosis, prognosis, or a therapeutic strategy. xAI introduces transparency and causality to these tools.

“Much of the current artificial intelligence in healthcare is primarily black box: algorithms make a prediction, but they don’t explain why,” said Dr. Taylor, Chairman and Co-founder of SpIntellx and the Director of the University of Pittsburgh Drug Discovery Institute. “With xAI, the algorithms make their prediction and inform the pathologist or the oncologist why they made that prediction. Our team has led the pack in the implementation of xAI in biomedical sciences, specifically in pathology.”

HistoMaprTM: A computational guide for pathologists

HistoMapr has been developed for analysis of microscopic structures across cross sections of pathology samples. It quickly identifies regions of interest of diagnostic significance. By using xAI, HistoMapr can explain why these regions are considered as high-risk lesions, and whether the patient is at risk of developing breast cancer or not. Currently, HistoMapr is used as a guide or training tool to aid pathologists’ decision-making in cancer diagnosis.

HistoMapr has provided 56% faster case review and more accurate (83% vs. 52%) diagnosis of high-risk lesions, which are considered as challenging cases in pathology, in a pilot study on breast tissue biopsies at Magee-Womens Hospital.

TumorMaprTM: A tool to improve personalized therapeutic strategies, diagnostic, and prognostic tests for cancer

“TumorMapr turns [third party] data into knowledge, identifying tumor hotspots with cancer activity, or regions of interests, and then infers the network biology that is driving these regions to be hotspots by applying spatial intelligence on multiplexed and highly multiplexed (hyperplexed) fluorescence images,” said Dr. Chennubhotla.

TumorMapr has the ability to build networks and make connections between these selected hotspots that are positive for cancer biomarkers. The xAI feature in TumorMapr explains why a specific patient would not respond to a certain treatment or what components of the network can be targeted as a therapeutic strategy. This enables precision medicine, as it can have impact on prognosis and therapeutic strategies for each individual case.

To date, the SpIntellx team has filed 7 patent applications (2 PCT patents issued and 5 pending) and holds a software license through University of Pittsburgh’s Innovation Institute. They are continuing to pursue SBIR funding through the National Institute of Health (NIH) to further provide opportunities to run validation studies for use of TumorMapr in precision oncology.

Interested in learning more about SpIntellx, HistoMaprTM, and TumorMaprTM?  Visit the SciVelo blog.