The Challenging Aspects of Developing AI/ML-Based Diagnostic Pathology Platforms

Immunotherapy & Machine Learning: Where Are We in 2022?

Recent advances in immunotherapy, such as the use of checkpoint inhibitors, are transforming the treatment of many cancers. Yet, not all patients respond sufficiently and efforts are underway to identify and validate biomarkers capable of selecting the most suitable patients for these therapies. Detecting the presence of biomarkers in tissue by analyzing tissue samples and recognizing patterns within them, however, is a time-consuming task for pathologists and often results in less-than-optimal reproducibility.

However, a potential game-changing technology is on the horizon. Emerging Artificial Intelligence (AI) and machine learning (ML) based diagnostic pathology platforms using dynamic algorithms developed through deep neural networks have the potential to support pathologists and oncologists by improving the efficiency, reproducibility, accuracy and precision of their work. While no such dynamic diagnostic pathology platforms that rely on AI are nearing market registration as of yet, the technology is evolving rapidly and the day will come when it is used routinely in drug development, clinical trials and precision medicine. Not far in the future, the first application of this technology will be in selecting and stratifying patients for immunotherapy and have value in any therapeutic area that utilizes whole slide imaging (WSI) to determine the expression of a biomarker to inform patient treatment.

When properly developed algorithms can support the pathologist in increased efficiency and accuracy in tissue interpretation it will dramatically improve patient selection for treatment and patient outcomes. The hope is that in the future, drug developers will routinely use this technology to stratify patients into clinical trials to improve patient response. Further, physicians will eventually rely on this information to identify predictive biomarkers as standard of care inform treatment decisions.

The development pathway for this technology is largely unchartered and technology developers should seek the counsel and support of a development partner who is:

1. Immersed in the business of conducting immunotherapy and AI/ML-based device trials. The development partner should have strong, existing relationships with study sites to expedite site recruitment and study start up.

2. Possesses large-data set know-how. Data collection should be well understood operationally and the development partner should have experience managing and integrating large data sets.

3. Well-versed in regulatory technology product development. This partner will also understand the specific challenges technology development customers face, including interaction with the regulatory agency and the required regulatory pathway for AI/ML-based technology.

4. Specialized capabilities in managing prospective studies. Partnering with the right team with specialized knowledge of immuno-oncology, and digital pathology to support the collection of data required by developers to train algorithms. This approach provides access to multi-disciplinary experience in conducting Immuno-Oncology trials and in guiding developers through the approval process.

Clearly, there are also many challenging aspects of developing AI/ML-based diagnostic pathology platforms that have little to do with computer programming and data science. A significant portion of the process is the stock and trade of research development organizations that help to bring drugs and medical devices to market. The best solutions are likely to come from development partners who are familiar with clinical trials, digital pathology and immunotherapy coupled with the unique regulatory expertise and knowledge required for this emerging area of software medical device development.