Skip to main content
Premium Trial:

Request an Annual Quote

PathAI Leveraging Acquisition, $165M Financing to Expand Dx Business


NEW YORK – Flush with $165 million after a recent Series C round and with a new laboratory under its umbrella, digital pathology company PathAI is making strides to expand its diagnostics capabilities and drug development work.

The company has two main points of focus, CEO and cofounder Andy Beck said — supporting drug development through analyzing completed clinical trials for biomarkers and providing diagnostic algorithms for different diseases.

Earlier this week, it acquired Poplar Healthcare Management, the management service organization for Poplar Healthcare, which offers a variety of laboratory testing services, to help it broaden its diagnostics business.

Most of the Boston-based firm's data, thus far, has been from its work on drug development and measuring treatment response, although working with Poplar will allow it to further develop its diagnostic capabilities, Beck said.

PathAI analyzes clinical trial data via hematoxylin and eosin-stained and immunohistochemistry-stained slides in a variety of therapeutic areas, namely oncology and nonalcoholic steatohepatitis, or NASH, Beck said.

In the past year, it has extended its yet-to-be named platform to incorporate its disease-specific algorithms within prospective clinical trials and partner with biopharmaceutical companies on regulated devices, including in vitro diagnostics and companion diagnostics to determine the efficacy of different treatments.

The firm's algorithms can be used in clinical trials for quality control, biomarker assessment, tools for patient enrollment, and candidate companion diagnostics, Beck said. It currently is involved in a number of ongoing clinical trials across all phases, he added.

Earlier this year, the company presented data at the American Society of Clinical Oncology Virtual Scientific Program 2021 with Roche's Genentech validating its artificial intelligence-powered system to determine a patient's response to treatment in non-small cell lung cancer.

The machine learning-based PathR algorithm from PathAI was applied to data from a Genentech LCMC3 trial and showed that model calculations of pathologic response of resection specimens after neoadjuvant atezolizumab (Genentech's Tecentriq) in the cancer were robust.

Researchers found "that automation may provide a scalable alternative to or complementary tool for manual assessment," they wrote in the ASCO abstract.

Pathologic response to this treatment is currently calculated manually from tumor resection slides by pathologists, and the response can be used as an efficacy endpoint in clinical trials looking at neoadjuvant therapies for patients with resectable NSCLC, PathAI said in a statement.

According to the data, the machine learning models were at least as accurate as manual assessment, but further data maturation and validation is necessary to understand their effectiveness more thoroughly.

"The work we've presented demonstrates the utility of using an AI-based approach to support an accurate, reproducible, and efficient way of measuring treatment response across a variety of diseases in both oncology and non-oncology applications," Beck said.

Rohit Loomba, a professor of medicine and director of hepatology at the University of California, San Diego, who worked with PathAI on a study to determine its use case in NASH, has served as an investigator on previous NASH trials that have looked at "the role of AI and quantitative automated reads on liver histology to assess treatment response," he said via email.

Loomba said that the PathAI data is emerging and may "provide a new way [to] look at liver disease assessment by eliminating the subjective assessment and making it much more objective and more accurate [by] more precisely quantifying treatment response."

In addition, in 2019, PathAI presented retrospective data together with Bristol Myers Squibb that found its algorithm was comparable to human pathologists for scoring PD-L1 expression on tumor and immune cells for predicting immunotherapy response in lung and bladder cancer.

Tumor scoring, input, and output for the digital algorithm are for measuring a patient's response to treatment, as is a manual assessment by a pathologist, but with the AI-based method, the goal is for the reproducibility and accuracy of the method to be higher, Beck said. "The input's the same and the output's the same, it's more that the processing is very different."

The goal for pathologic response is to quantify the amount of viable tumor following treatment, which constitutes a percentage of the total tumor bed — in other words, how much of the tumor died, Beck said.

"The power of the AI system is you can actually train the algorithm to identify the pattern of viable tumor, and then just deploy it to exhaustively analyze every slide of the patient and to actually exactly count the area of tissue that's dead and the area of tissue that's still alive to compute that exact percentage in a totally reproducible, standardized fashion across slides and across patients," Beck said.

For a big case, manually reviewing 20 slides can take a pathologist more than half an hour, Beck said. Artificial intelligence solutions, however, can be trained to provide results in one minute, with multiple computations being done in parallel, analyzing slides at the same time.

In the AI workflow, slides go into a whole-slide imaging system and are converted into large digital files, which become the input into PathAI's algorithms. Every pixel of the image is predicted and the overall predictions on an entire slide are summarized for the pathologist to sign off on if he or she agrees with the diagnosis.

When developing an algorithm for a specific disease, the company first determines the input and output and thinks about how it will be validated. It then puts together the datasets, often slides annotated by board-certified pathologists, and sends the data to machine learning engineers to build the algorithm. The algorithms and their clinical workflow are then validated on external data to ensure efficacy and safety, Beck said.

Using this technology also means slides don't have to be mailed to pathologists, cutting down on the turnaround time for results, Beck said. It allows for the "digitization and distribution of work," he added.

The firm's focus since it started has been on processing whole-slide images, but it has invested in its platform to make it "so robust" that it can now "rapidly develop, validate, and deploy new models," Beck said.

The regulatory process for PathAI's algorithms depends on the specific use case and indications of the algorithms, although it is typically part of a larger medical device, Beck said. The US Food and Drug Administration has approved a number of image analysis algorithms for use in pathology that the company can use as references, and the process for getting its products cleared is similar to the IVD process, he said.

While the company has several products in submission with the FDA, none of its products have been approved for marketing. Beck said PathAI is working with the agency on using its tool to "support pathologists for both patient enrollment as well as assessment of therapeutic response" in trials for NASH.

Diagnostic progress

The underlying task is the same for clinical diagnosis and can be deployed in the clinical setting where a clinician is attempting to determine whether a patient has cancer or not, Beck said.

"We're incredibly excited about this ability to really accelerate the field of pathology," Beck said.

The algorithms can be used to develop companion diagnostic tests as well, allowing for better selection of patients who will be the best responders to therapies or patients who will likely not respond to therapy, Beck said.

Earlier this year, PathAI broadened an existing partnership with Laboratory Corporation of America to include the potential development of companion diagnostics using the firm's algorithms.

The focus will be on using the algorithms in programs managed by Labcorp Drug Development in prospective clinical trials, specifically their use in the development of companion diagnostics.

Meantime, its acquisition of Poplar Healthcare Management will help it step further into the diagnostic space. PathAI will partner with the CLIA-certified anatomic pathology laboratory to develop and validate AI-powered algorithms to be used for specific diseases with a focus on accuracy and reproducibility, Beck said. Poplar will distribute the diagnostic tests, he added.

While PathAI is in the early stages of developing and validating these tests specifically with the clinical workflow in mind, it plans to produce a "broad portfolio of algorithms that will cover the vast majority of slide types [and] specimen types being developed within the pathology lab," Beck said.

The firm is starting with high volume specimen types, such as H&E-stained slides and IHC slides, and major disease areas like dermatopathology, prostate pathology, and gastrointestinal pathology. Poplar currently offers gastrointestinal and dermatology tests, along with oncology, urology, and women's health services.

"We want to have one foot in the diagnostics side and one foot in the drug development side and think we can really provide the most value to patients through connecting these two worlds," Beck said.