Last year, venture capitalists invested significant sums in startups developing artificial intelligence (AI) and machine-learning technologies for medical image analysis and support, with some groups projecting a $20-billion market by 2031, according to STAT News.

Yet, providers and pharmaceutical companies that could benefit from the power of these technologies remain wary due to perceived shortcomings and inflated hype around their use for clinical and clinical research purposes. They are well aware of companies’ mixed results with introducing AI-driven clinical decision support systems and data management into medicine, notably the disappointing effort by the high-profile IBM Watson Health, resulting in the sale of that business to a private equity firm earlier this year.

Skeptics notwithstanding, much has changed in recent years. Today, the field of AI in healthcare is at an inflection point. Innovative machine learning and computer vision algorithms, when based on high-quality inputs, are creating a new layer of information and enabling development of non-linear multi-modal models to better understand complex disease states. Applied in the pharmaceutical industry setting, these technological advances have the potential to streamline the highly inefficient drug development process and improve access to quality clinical care, bringing precision medicine closer to reality for patients.

Several trends are prompting this dramatic transition. First and foremost, vendors increasingly recognize that their value lies in pursuing solutions that address well-defined use cases, in contrast to earlier attempts to apply AI generically across the board to healthcare. Further, a growing body of published research describes consistent methods by which the clinical applications of AI are assessed and validated, mitigating concerns about the capabilities and safety of AI in healthcare. These developments are at the heart of the growing clinical acceptance of AI-driven products and services.

Computational innovations are also motivating the life sciences industry to embrace AI. Computing power has significantly strengthened in the past decade and is faster, more intuitive, and cost efficient. Cloud infrastructure is increasingly robust and inexpensive, enabling massive amounts of data to be stored and broadly accessible to industry users. As an example, TensorFlow, the open-source platform for machine learning and AI, which Google originally developed for internal use, is now incorporated into the Google Cloud Platform, which Google distributes widely to the public. Image quality is also improving, due to inexpensive storage capacity and growing availability of advances such as 4K resolution. As a result, datasets are more robust and capable of delivering higher quality annotations and better fidelity, which ensures AI-driven conclusions have greater accuracy.

Practitioners will benefit from these technological advances in exciting ways. Physicians perform many procedures that require them to examine images for long periods of time, leading to subjective conclusions. This is true in endoscopy, for example, which relies on a highly skilled physician-dependent procedure, in which large amounts of information are generated and rigorously interpreted via close examination of mucosa.

Endoscopists have high-resolution imaging equipment available in their offices, as well as enough bandwidth to record and store single-unit images from procedures. More recently, they are also benefiting from advances in cloud computing, which enable them to store and transmit full-length videos of entire procedures—quickly, inexpensively and in large volumes. This capability, in turn, has the potential to facilitate extraction of highly granular data that can be used to train algorithms, which ultimately could enhance drug development and clinical decision making.

AI Use Case: Enhancing Clinical Trial Recruitment in Inflammatory Bowel Disease

Gastrointestinal endoscopy is an example of a field that is poised to benefit from the power of AI to improve patient care, drug development and clinical workflows. For several reasons, GI practitioners have seen fewer advances from precision medicine technologies compared to those who treat other serious illnesses, such as oncologists and infectious disease specialists.

Early on, Iterative Scopes identified the subjective nature of endoscopic readings as a challenge both for drug developers and clinicians, and recognized the potential of AI to transform gastroenterology, when applied to specific use cases. We chose to prioritize inflammatory bowel disease (IBD), a chronic, debilitating, and common condition, which affects more than 3 million people in the United States alone. We identified clinical trial recruitment as a key hurdle for pharmaceutical companies and clinical researchers working on improving drug development for IBD.

The variability and subjectivity of endoscopic disease scoring used by referring physicians to identify patients who are eligible for IBD clinical trials creates a time-consuming challenge for both the referring clinicians and pharmaceutical companies developing new drugs. Iterative Scopes believes that AI could be used to improve the enrollment process for IBD clinical trials, and ultimately to build entirely new endpoints for evaluating drugs in effectiveness in IBD drug development. Our computational software automates interpretation of colonoscopy images and videos, enabling clinical trial investigators to arrive at standardized endoscopic disease scores for individual patients.

That said, the sophisticated computational powers needed to extract, manage, and interpret data are critical to successful application of AI, but insufficient on their own to change the paradigm of care and are increasingly commoditized. Companies with the resources and expertise can now create high-quality annotated training datasets from widely available, off-the-shelf hardware. While the quality of datasets that are the foundation for training algorithms has come under scrutiny, these challenges can be addressed through compliance with well-established industry protocols for validating AI models and the dedication of appropriate resources.

More challenging is combining the collective components needed to gain the trust of stakeholders who contribute to, and also benefit from, AI-driven advances in clinical trial design. Companies that take this kind of holistic approach to improving the drug development process will be able to differentiate themselves. Iterative Scopes, for example, has made it a priority to form partnerships with leading GI organizations, providing unparalleled access to robust, curated datasets, which can power the quality and scope of its AI-driven software.

At the same time, pharmaceutical companies are understandably reluctant to share their valuable, proprietary R&D data. But their input is critical, partly because that information is necessary to inform the training datasets that are the foundation of accurate algorithms. However, this approach requires a shift in traditional thinking about clinical trials and how to bring drugs to market. Inability to acknowledge these links could delay the development of more informative clinical trials, and raises questions about the most effective ways to harness the data necessary to inform algorithmic models.

AI and computer vision-driven algorithms can bring objectivity and speed to the drug development process by offering pharmaceutical sponsors more precise, quantifiable (and therefore objective) clinical endpoints by which to evaluate drug candidates. The ability to assemble a comprehensive ecosystem that enables development of high-quality AI-driven solutions to key challenges in drug development will differentiate successful players in this field.

Brice Wu is CPO and SVP Engineering at Iterative Scopes, a Cambridge, MA-based company that was spun out of MIT in 2017 and is a pioneer in the application of AI/ ML-based precision medicine to gastroenterology.

Previous articleEnzyme Study May Lead to Applications in Novel Drug Design and the Creation of Artificial Enzymes
Next articleNovel Combination Therapy Developed against Liver Cancer