The path from research to clinical use

Stratipath is a spin-out from Karolinska Institutet in Stockholm, Sweden, bringing pioneering research in AI and precision medicine into clinical use. The company was founded in 2019 by Johan Hartman, MD, PhD, Mattias Rantalainen, PhD, and Fredrik Wetterhall with a mission to radically improve cancer treatment decisions and patient outcomes.

Stratipath is a spin-out from Karolinska Institutet in Stockholm, Sweden, bringing pioneering research in AI and precision medicine into clinical use. The company was founded in 2019 by Johan Hartman, MD, PhD, Mattias Rantalainen, PhD, and Fredrik Wetterhall with a mission to radically improve cancer treatment decisions and patient outcomes.

The underlying problems in breast cancer diagnostics

The academic research field in AI-based image analysis, or “computational pathology”, is rapidly progressing, with the potential to revolutionise cancer diagnosis. AI-based systems are already able to identify cancer cells and classify tissue changes, as well as predict mutations and gene expression directly from routine digital tissue sections 1 2 3 4 5.

At the same time, pathology is facing a shortage of personnel and an increasing number of tissue samples that require advanced diagnostics, while policymakers and patients demand quick and accurate diagnoses. Unfortunately, investigations show a lack of quality and large regional discrepancies in diagnostics 6. For example, in breast cancer care, where tumour biology determines the risk of recurrence, a correct assessment of biomarkers from pathology results is essential to patient treatment. There are large differences in diagnoses, both between and within hospitals, leading to unequal care depending on where the patient lives.

The aggressiveness of breast cancer is largely determined by its tumour grade, which is analysed by the pathologist through a time-consuming and careful procedure involving the examination of multiple parameters. Particularly for patients with oestrogen receptor-expressing tumours (ER+), tumour grade plays an important role in treatment recommendations. Patients with tumour grade 1 have a low risk of lymph node metastasis and relapse, while those with tumour grade 3 have a significantly worse prognosis 7 8. Over 50% of all tumours fall into an intermediate category, tumour grade 2 6. For patients with grade 2 tumours, this translates to an intermediate risk, information which has limited clinical value. Researchers have previously demonstrated that grade 2 breast cancer show gene expression patterns that are identical to those in either grade 1 or grade 3 tumours 9 10.

Analyses based on gene expression profiling are today used as decision support in clinical routine, but they are expensive and require the purchase of special equipment or samples to be sent for external analysis. Additionally, the treating physician may have to wait for one to two weeks for analysis results. Overall, the commercial regulatory approved gene expression profiling tests used for decision support in prognostic evaluations of breast cancer are not accessible to all patients due to their expense and long wait times for results.


The academic research team

To address these problems, the research groups of Johan Hartman and Mattias Rantalainen at Karolinska Institutet, teamed up to develop an AI model, which would divide grade 2 breast tumours into low and high risk groups based on microscopy images. By leveraging the immense amount of information contained in microscopy images, the research team aimed to directly distinguish these two groups based on routinely stained microscopy images. The research team has constructed the necessary capacity and computational power to develop their own AI-based image analysis systems for the pathology of the future.


How the AI system was trained and validated

AI systems based on deep learning, are trained using large data sets, such as microscopy images, which are first optimised for quality prior to analysis by AI models. To train the AI system to identify cancer in microscopy images, tumour areas were first marked by experienced pathologists in a substantial number of images. The trained AI system could then automatically detect cancer in new images and serve as an aid in further analysis of the detected cancer area. Using AI models, the research team at Karolinska Institutet was able to teach the system to recognise the patterns and structures that represent grades 1 and 3 breast tumours. When the AI models were applied to grade 2 tumours, they could be divided into high and low risk groups, with similar survival profiles to grade 3 and grade 1 tumours, respectively. Intriguingly, the two groups exhibited very similar gene expression profiles as grade 3 and 1, respectively. The AI model was validated in independent breast cancer cohorts and exhibited a prognostic strength similar to commercial gene expression profiles 11.


The steps towards a product for clinical use

In the light of the clinical need for rapid, accurate, and prognostic analyses for ER+ and HER2-negative breast cancer patients, the research team at Karolinska Institutet embarked on a mission to render academic research accessible to patients. With the aim of providing all breast cancer patients with equal access to diagnostics, the team began the development of a product for clinical use. The company Stratipath was founded in 2019 by Johan Hartman, Mattias Rantalainen and Fredrik Wetterhall, with the support of Karolinska Innovations and Vinnova 12. The development process is both costly and time-consuming, requiring regulatory, medical, and technical expertise.

After undergoing a series of quality and safety tests, Stratipath Breast, a first-of-its-kind AI-based image analysis solution for decision support in breast cancer diagnostics, was CE-IVD marked, allowing clinical use.


The future

Commercial gene expression analyses are currently used for risk stratification of breast cancer, yet expensive, not widely available, and time-consuming. AI-based image analysis offers many advantages to current methods. For example, image analysis can analyse multiple areas of a tumour, allowing for the detection of heterogeneity and spatial differences in aggressiveness, and can automatically detect cancer or premalignant changes, as well as biomarkers. This provides valuable decision support to both pathologists and treating physicians.

Stratipath will continue to develop AI-based image analysis solutions for other purposes and cancer diagnoses, with the goal of providing cost-effective precision medicine to a greater number of patients.

1↑ Cruz-Roa A, Gilmore H, Basavanhally A, Feldman M, Ganesan S, Shih NNC, et al. Accurate and reproducible invasive breast cancer detection in whole-slide images: A Deep Learning approach for quantifying tumor extent. Sci Rep. 2017 Apr;7:46450.
2↑ Bulten W, Pinckaers H, van Boven H, Vink R, de Bel T, van Ginneken B, et al. Automated deep-learning system for Gleason grading of prostate cancer using biopsies: a diagnostic study. Lancet Oncol. 2020 Feb;21(2):233–41.
3↑ Campanella G, Hanna MG, Geneslaw L, Miraflor A, Werneck Krauss Silva V, Busam KJ, et al. Clinical-grade computational pathology using weakly supervised deep learning on whole slide images. Nat Med. 2019 Aug;25(8):1301–9.
4↑ Fu Y, Jung AW, Torne RV, Gonzalez S, Vöhringer H, Shmatko A, et al. Pan-cancer computational histopathology reveals mutations, tumor composition and prognosis. Nature Cancer. 2020 Jul 27;1(8):800–10.
5↑ Kather JN, Heij LR, Grabsch HI, Loeffler C, Echle A, Muti HS, et al. Pan-cancer image-based detection of clinically actionable genetic alterations. Nature Cancer. 2020 Jul 27;1(8):789–99.
6↑ Acs B, Fredriksson I, Rönnlund C, Hagerling C, Ehinger A, Kovács A, et al. Variability in Breast Cancer Biomarker Assessment and the Effect on Oncological Treatment Decisions: A Nationwide 5-Year Population-Based Study. Cancers. 2021;13(5):1166.
7↑ Rakha EA, El-Sayed ME, Lee AH, Elston CW, Grainge MJ, Hodi Z, et al. Prognostic significance of Nottingham histologic grade in invasive breast carcinoma. J Clin Oncol. 2008;26(19):3153-8.
8↑ Rakha EA, Reis-Filho JS, Baehner F, Dabbs DJ, Decker T, Eusebi V, et al. Breast cancer prognostic classification in the molecular era: the role of histological grade. Breast Cancer Res. 2010;12(4):207.
9↑ Sotiriou C, Wirapati P, Loi S, Harris A, Fox S, Smeds J, et al. Gene expression profiling in breast cancer: understanding the molecular basis of histologic grade to improve prognosis. J Natl Cancer Inst. 2006;98(4):262–72.
10↑ Ivshina AV, George J, Senko O, Mow B, Putti TC, Smeds J, et al. Genetic reclassification of histologic grade delineates new clinical subtypes of breast cancer. Cancer Res. 2006;66(21):10292–301.
11↑ Wang Y, Acs B, Robertson S, Liu B, Solorzano L, Wählby C, et al. Improved breast cancer histological grading using deep learning. Ann Oncol. 2022 Jan;33(1):89–98.
12↑ Vinnova. Innovationsmiljöer för mer träffsäkra lösningar inom hälsa [Internet]. 2021 [cited 2022 Aug 4]. Available from: