Artificial intelligence (AI) has growing appeal in the field of drug discovery and development, offering the ability to process more information faster and possibly with less margin of error than human efforts. While many companies are applying this technology in pursuit of new drug targets for disease, some are looking to its utility in diagnosing those diseases. The hope is that with the aid of high-throughput technology scanning and comparing hundreds or thousands of cases for similarities and outcomes, disease states and progression can be better predicted and managed.
Dascena Inc. is one such firm betting on that approach, pursuing early disease intervention through better and earlier diagnosis. The company, which recently emerged from stealth mode, announced in the second quarter of 2020 that it had closed a $50-million Series B financing led by Frazier Healthcare Partners, with participation from Longitude Capital, existing investor Euclidean Capital and an undisclosed investor.
Dascena plans to use the funds to advance its portfolio of machine-learning algorithms meant to inform patient care strategies and improve outcomes.
“At Dascena, we believe in the power of machine learning to improve patient care and outcomes, and we continue to develop algorithms to do just that,” said Ritankar Das, founder and CEO of Dascena. “Machine learning is transforming how we solve problems across industries, and by applying this technology to healthcare, we will enhance the quality and efficiency of patient care. Our flagship sepsis algorithm, InSight, has produced significant decreases in mortality and hospital length of stay among patients with sepsis.”
According to Dascena, its algorithms have been validated in several studies, with data published in 18 peer-reviewed publications. One such algorithm, InSight, led to a 58-percent reduction in patient mortality and a 21-percent reduction in length of hospital stay in a randomized, controlled trial of hospitalized patients in the ICU in 2017.
Dascena further validated these data with a second, larger study, having announced the results on April 30. All told, 17,758 patient encounters met two or more systemic inflammatory response syndrome criteria and were included in this analysis. InSight intervention recommendations led to a 39.5-percent reduction in mortality, a 32.27-percent reduction in length of hospital stay and a 22.74-percent reduction in 30-day readmission.
Cancer is another field where there are high hopes for AI’s diagnostic potential. Many types of cancer, such as pancreatic or colon cancer, don’t manifest symptoms until later in the course of the disease, at which point a patient’s survival outlook is very poor. Cancer diagnostics that screen for specific genetic mutations have seen increased usage, but as several tumor types are heterogeneous and can present with a variety of mutations, some patients can fall through the cracks.
To try and address this issue, a new approach developed by a University of Waterloo-led team is seeking to applying AI in detecting cancer in digital images of tissue samples. Digital images of biopsies from confirmed cancer cases were searched with AI to be compared against digital images of biopsies from suspected cases.
This study was conducted over the course of four months on an archive provided by the National Cancer Institute, part of the U.S. National Institutes of Health, and was part of a five-year project supported by the Ontario government. Their results were presented in a paper titled “Pan-Cancer Diagnostic Consensus Through Searching Archival Histopathology Images Using Artificial Intelligence,” which was published in Nature Digital Medicine.
The archive in question boasts roughly 30,000 digitized slides from nearly 11,000 patients, making it the largest publicly available archive in the world. The researchers found that the AI technology was capable of reaching up to 100-percent accuracy for 32 types of cancer in 25 different organs and body parts.
“AI can help us tap into our medical wisdom, which at the moment is just sitting in archives,” commented Dr. Hamid Tizhoosh, director of the Laboratory for Knowledge Inference in Medical Image Analysis (KIMIA Lab) at Waterloo. “We showed it is possible using this approach to get incredibly encouraging results if you have access to a large archive. It is like putting many, many pathologists in a virtual room together and having them reach consensus.”
And the use of AI in diagnostics—and indeed, in the industry in general—isn’t likely to be a passing fad, but a new standard. Pistoia Alliance, a not-for-profit life-sciences organization, recently shared its 2030 Life Sciences and Health Go Digital report regarding AI and machine-learning in the pharmaceutical and life sciences field.
The report notes that as of 2018, “more than a third of healthcare providers had made investments into healthcare AI and medical predictive analytics, preparing for the next generation of automated healthcare. For example, in pathology, the extensive use of whole-slide imaging aligned with pattern recognition methods based on deep learning, as well as incorporating clinical, radiologic and genomic data, allowed highly sophisticated, rapid and accurate diagnosis and prognosis.” It also cites how using AI to analyze electronic medical records with a focus on new prescriptions can catch physician errors and prevent issues such as overdose or other adverse events.
According to John Wise, co-author of the report, it won’t be a fully digital future, but one where AI does the heavy lifting before human experts confirm the results. Allie Nawrat, a pharmaceutical writer for GlobalData, noted that “The report predicts that diagnosis in 2030 will be largely determined and improved by artificial intelligence (AI) ... AI’s increasing role in diagnosis may shift the role of physicians into being more like patient educators, responding to what the AI systems conclude.”