This story has been updated to correct information related to Aiforia's funding.
CHICAGO – Following a series of CE-IVD marks, an initial public offering, and a key partnership with the Mayo Clinic, Finnish bioinformatics company Aiforia Technologies is accelerating its efforts to automate pathology with artificial intelligence.
In June, the Helsinki-based company announced the launch of the CE-marked Aiforia Clinical Suite Viewer, which provides clinical decision support to pathologists by applying artificial intelligence to digital histopathological slides. At the time, Aiforia called that CE mark a "key advancement" toward its goal of offering end-to-end clinical pathology workflow automation.
The firm also has five CE-IVD marked AI models, its most recent being for the Aiforia Clinical AI Model for Breast Cancer; PR (for "progesterone receptors") that automates the identification of tumor areas and the calculation of progesterone receptor-positive and PR-negative cells from whole-slide images. Other marks are for models that assist in breast, lung, and prostate cancer diagnostics based on biomarkers including Ki-67 and PD-L1.
Aiforia software tools deliver to pathologists reports containing quantitative analysis to assist physicians in the diagnostic process.
"Our strategy is really to bring a good portfolio of different AI models" to market, CEO Jukka Tapaninen said. "Our idea is really to build … 10 to 20 AI models [for common cancers] so that we start covering about 80 percent of the clinical workflow."
Tapaninen said that Aiforia has a HER-2 AI model for breast cancer that is all but ready for market, but it does not yet have a CE-IVD mark. The company sought all of its current CE marks before the European Union's new In Vitro Diagnostic Regulation (IVDR) took effect May 26. Tapaninen said this new regulatory environment has caused Aiforia to hold off on applying for CE marking on additional products until it can be sure it is in compliance with IVDR, likely before the end of the year.
According to Tapaninen, the company is not yet seeking US Food and Drug Administration clearance, largely because most of the company's customers in the US are building their own laboratory-developed tests on the Aiforia platform. Tapaninen said that customers have built about 400 of their own AI models for research purposes through a development platform called Aiforia Create that predates the company's ready-made AI models.
The platform's annotation assistant is patented in the US.
"We enable the pathologists to make their own AI models" to annotate test results. Tapaninen explained. "They don't need to have technical skills."
Aiforia incorporated in 2013 and began operations a year later as a spinout from the Finnish Institute for Molecular Medicine (FIMM). The company was originally called Fimmic and sought to provide a way to share pathology images in a cloud environment. The Aiforia name stands for "artificial intelligence for image analysis," according to Tapaninen.
The company was founded by Kari Pitkanen, who remains a project director of translational genomics at FIMM, as well as by Johan Lundin and Mikael Lundin from the Karolinska Institute in Sweden.
Tapaninen joined in 2015 as Aiforia began to apply deep learning-based AI to analyze images on a cloud called WebMicroscope. The firm rebranded the platform as Aiforia Cloud in 2018.
Aiforia had an initial public offering in December 2021 that raised about €30 million ($29.4 million) and is traded on the Helsinki-based Nasdaq First North Growth Market. Tapaninen said that the company got lucky with the timing, as the IPO market started to cool off in early 2022.
"We have a nice amount now of cash in our pockets so that we can really invest on speeding up the commercialization and the product development," Tapaninen said. The firm reported €32.3 million in cash and equivalents as of June 30.
Prior to the IPO, the firm raised about €28 million in venture capital, most recently in a €17.5 million Series B round that closed in September 2021. Aiforia also received a €2.1 million grant in 2019 from the EU's Horizon 2020 program to support development of deep learning AI models for primary cancer diagnosis.
The company's primary markets are Europe and the US, though Aiforia does count the University of Sydney as a customer of its pathology cloud platform. Tapaninen said that the firm is likely to generate more revenue from the US than any other market this year.
Some of that US revenue will result from the company's deal with Mayo, which in January announced a partnership with Aiforia to develop AI-powered pathology research support computing architecture.
The Mayo Clinic Department of Laboratory Medicine and Pathology is installing Aiforia software at various sites across the US to support remote collaboration among pathologists, who will also have the opportunity to develop their own AI models for research-focused image analysis with the help of Aiforia Create.
Tapaninen said that Mayo generates 2.6 million pathology slides a year, a volume that is difficult to manage with manual processes. Mayo has already validated the Ki-67 AI module and is actively using that technology, he added.
Other key US users include the Massachusetts Institute of Technology, Wake Forest University, and City of Hope National Medical Center.
Cancer geneticist and immunologist Peter Westcott, a newly appointed faculty member at Cold Spring Harbor Laboratory, worked with Aiforia Create when he was a postdoctoral fellow in the lab of Tyler Jacks at MIT's Koch Institute of Integrative Cancer Research.
There, Westcott developed an algorithm that he said is able to recognize non-small cell lung cancer tumors in mice and "really accurately grade those tumors" on the traditional scale of grades I to IV. "Now that's the routine analysis methodology used by the lab for all these lung cancer models to detect overall tumor burden," Westcott said.
He and his colleagues at MIT also developed a qualification algorithm for immunohistochemistry protocol to assure consistent, accurate grading of tumors based on CD4, CD8, and regulatory T-cells.
"There's this kind of game of Telephone, if you will, where there was this expert mouse pathologist [who] trained some of us on how to grade tumors," Westcott recalled. Then postdoc and graduate students would teach lab technicians based on what the pathologist taught them.
"You'd get this degradation in quality of grading to the point where there's some just outright, completely wrong grading that was going on in the lab," Westcott said. "We felt like there needed to be some sort of standardization that takes out the user bias in this metric of tumor grading."
Westcott expects to continue to use the technology at Cold Spring Harbor, though he said he will be evaluating Aiforia against at least one other software option.
He noted that while he was working with Aiforia during his time at MIT, he never did get a chance to complete a plan to integrate the tumor-grading algorithm with the T-cell IHC algorithm to study whether certain stages of tumor development are more associated with particular types of T-cell infiltration than others.
"I think if you could do good tissue registration, you could start layering on all these different algorithms that you're developing and then really ask complex questions," he said. "I think that would be really powerful."