The Finnish company Aiforia has developed a groundbreaking AI application capable of accurately and rapidly detecting changes, such as cancerous cells, in tissue sample images. This innovative collaboration between AI and human expertise significantly enhances image analysis, which previously relied solely on sensory observations. AI enables precise treatment plans and better research outcomes.
Aiforia's technology is used globally in research, drug development, and diagnostics. AI can analyze large amounts of data quickly and consistently, reducing the possibility for human error. Although AI does not replace human expertise, it provides valuable support for pathologists and scientists, and improves the accuracy of diagnoses and research. For example, analyzing vast tissue sample images is a labor-intensive task for humans, while AI can quickly identify suspicious areas, allowing doctors or researchers to focus on those concerning regions.
Aiforia originated from the University of Helsinki and the Institute for Molecular Medicine Finland (FIMM). The Lundin brothers, Mikael and Johan, initially developed a cloud-based tool for managing and sharing large tissue sample images. The third co-founder, Kari Pitkänen, brought extensive entrepreneurial experience from the biotechnology sector. Together, they decided in 2013 to commercialize their invention, giving birth to Aiforia. The company started with a five-member team and a product ready for the market from the beginning. Simultaneously, the company continued developing AI-based image analysis technology, and in 2018, they launched a new product—the first of its kind in the world. In its early years, Aiforia’s clients were research groups and institutions worldwide, later expanding to pharmaceutical companies and clinical diagnostics providers.
From Cancer Cells to Concrete Quality
Aiforia's AI models are based on deep learning neural networks, which can recognize
complex patterns and features in medical images.
"Fortunately, we chose deep learning neural networks and convolutional neural networks as our technology base," notes the company's COO, Kaisa Helminen.
These technologies have since become the dominant technology in various applications, such as facial recognition as an example. Aiforia's technology has even been used for monitoring concrete quality. Users can train the AI developed by Aiforia to recognize virtually any image data. Aiforia's technology was developed as cloud-based from the outset, which has contributed to the company's success. Kaisa Helminen emphasizes that Finland has a long tradition in this field. "We were also lucky in the beginning, as we found top experts in both image analysis and software development." The company's location in Finland has also been a competitive advantage, as competition for top talent is much fiercer in places like Silicon Valley.
Exponential Growth of AI – Humans at the Helm
Aiforia's models have been trained on vast amounts of data, enabling their ability to identify pathological changes and other significant medical features. The learning process is guided by humans.
“Humans teach AI what an immune cell is and what a tumor looks like. Our software is specifically designed for easy and quick learning, allowing experts to transfer their knowledge to AI,” says Kaisa Helminen.
Once trained, AI can repeatedly perform the taught task with precision and consistency. The expert verifies the findings.This approach saves experts considerable time and tedious calculations. Additionally, AI does not tire, unlike humans, who may make mistakes when exhausted or under pressure. The results produced by AI are visualized on the image sample, making the tool easy and intuitive to use. Experts can either accept the results or adjust them if they disagree with the AI. However, it is essential to remember that AI can only perform what it has been taught—meaning the human contribution remains indispensable. For example, each type of cancer is unique, and its specific features must be individually taught to the AI. Currently, Aiforia has regulatory-approved solutions for diagnosing breast, prostate, and lung cancers, and the company is continuously developing more AI models for various diseases. Aiforia's customers have already trained AI models for over a thousand applications.
Predicting Treatment Response
Aiforia’s AI can also be used alongside other clinical data to predict treatment outcomes, enabling significant cost savings. In a study conducted with the Mayo Clinic in the United States, the AI accurately predicted the non-recurrence of colorectal cancer, potentially sparing patients from unnecessary chemotherapy. Some estimates suggest that up to one-third of administered treatments are unnecessary.(*)
Applications offering more accurate predictions could save hundreds of millions of dollars in the United States alone.
In the future, the development of generative AI and image analysis is expected to advance rapidly. New innovations, such as more advanced deep learning models and real-time image recognition, will improve the accuracy and speed of diagnoses. Aiforia sees AI’s role growing particularly in understanding diseases and research, which could help in developing new treatments. It is also possible that in the future, AI models will independently handle simple tasks like screenings, allowing experts to focus only on samples with findings. AI also enables the analysis of significantly larger tissue samples, which could contribute to the development of entirely new diagnostic criteria.
(*) Additional information on colorectal cancer research:
Quantitative Pathologic Analysis of Digitized Images of Colorectal Carcinoma Improves Prediction of Recurrence-Free Survival. Pai, R K. et al. Gastroenterology, Volume 163, Issue 6, 1531–1546.e8 (2022)
Development and initial validation of a deep learning algorithm to quantify histological features in colorectal carcinoma including tumor budding/poorly differentiated clusters. Pai R K et al. Histopathology 79, Issue 3, 391–405 (2021)
Improved Risk-Stratification Scheme for Mismatch-Repair Proficient Stage II Colorectal Cancers Using the Digital Pathology Biomarker QuantCRC. Wu et al. Clinical Cancer Research, Epub ahead of print. doi: 10.1158/1078-0432.CCR-23-3211 (2024)
Comments