Artificial Intelligence Utilized to Identify Early Disease Indicators in Blood Tests

This article marks the third installment in a six-part series exploring the transformative impact of artificial intelligence (AI) on medical research and therapeutic approaches. Audra Moran, who leads the Ovarian Cancer Research Alliance (Ocra), a global charitable organization based in New York, describes ovarian cancer as “rare, underfunded, and deadly”. As with all cancers, earlier detection significantly improves outcomes. Most ovarian cancer originates in the fallopian tubes, meaning it may have already metastasized by the time it reaches the ovaries. Ms. Moran states, “Five years prior to ever having a symptom is when you might have to detect ovarian cancer, to affect mortality.” However, novel blood tests are emerging that leverage artificial intelligence (AI) to identify indicators of this cancer at its nascent stages. Beyond cancer, AI also has the potential to accelerate other blood tests for potentially fatal infections, such as pneumonia. Dr. Daniel Heller, a biomedical engineer at Memorial Sloan Kettering Cancer Center in New York, and his team have developed a testing technology employing nanotubes—minute carbon tubes approximately 50,000 times smaller than the diameter of a human hair. Roughly two decades ago, scientists began discovering nanotubes capable of emitting fluorescent light. Over the past ten years, researchers have mastered modifying the properties of these nanotubes to enable them to react to nearly any substance in the blood. Currently, it is feasible to introduce millions of nanotubes into a blood sample, causing them to emit varying wavelengths of light depending on what adheres to them. Nevertheless, the challenge of interpreting this signal remained, a task Dr. Heller likens to matching a fingerprint. In this context, the “fingerprint” represents a pattern of molecules binding to sensors with differing sensitivities and binding strengths. Yet, these patterns are too subtle for human perception. He explains, “We can look at the data and we will not make sense of it at all,” adding, “We can only see the patterns that are different with AI.” To decipher the nanotube data, the information was fed into a machine-learning algorithm. The algorithm was then instructed to differentiate between samples from patients diagnosed with ovarian cancer and those from individuals without it. This training dataset included blood samples from people with other cancer types or gynecological conditions that could be mistaken for ovarian cancer. A significant hurdle in developing AI-driven blood tests for ovarian cancer research is the disease’s relative rarity, which restricts the available data for training algorithms. Furthermore, much of this existing data is confined within the hospitals where patients received treatment, with limited data sharing for researchers. Dr. Heller characterized the process of training the algorithm on data from only a few hundred patients as a “Hail Mary pass.” Despite this, he noted that the AI achieved greater accuracy than the best currently available cancer biomarkers, and this was merely the initial attempt. The system is currently undergoing further investigations to assess potential enhancements using larger sensor arrays and samples from a greater number of patients. Increased data can refine the algorithm, much like how self-driving car algorithms improve with more real-world testing. Dr. Heller expresses considerable optimism for the technology. He states, “What we’d like to do is triage all gynaecological disease – so when someone comes in with a complaint, can we give doctors a tool that quickly tells them it’s more likely to be a cancer or not, or this cancer than that.” Dr. Heller estimates this advancement could be “three to five years” away. AI’s utility extends beyond early detection; it is also accelerating other blood tests. For a cancer patient, contracting pneumonia can be fatal, and given that approximately 600 different organisms can cause pneumonia, doctors must perform multiple tests to pinpoint the infection. However, new blood test methodologies are streamlining and expediting this diagnostic process. Karius, a California-based company, employs artificial intelligence (AI) to help identify the precise pneumonia pathogen within 24 hours and recommend the appropriate antibiotic. Karius chief executive Alec Ford explains, “Before our test, a patient with pneumonia would have 15 to 20 different tests to identify their infection in just in their first week in hospital – that’s about $20,000 in testing.” Karius maintains a microbial DNA database containing tens of billions of data points. Patient test samples can be cross-referenced with this database to identify the exact pathogen. Mr. Ford asserts that such a capability would have been impossible without AI. A current challenge lies in the fact that researchers do not always fully comprehend all the correlations an AI might establish between test biomarkers and diseases. Over the past two years, Dr. Slavé Petrovski developed an AI platform named Milton, which has successfully identified 120 diseases with an accuracy exceeding 90% by utilizing biomarkers from the UK Biobank data. Discovering patterns within such vast datasets is a task uniquely suited for AI. Dr. Petrovski, a researcher at the pharmaceuticals giant AstraZeneca, notes, “These are often complex patterns, where there may not be one biomarker, but you have to take into consideration the whole pattern.” Dr. Heller applies a similar pattern-matching approach in his ovarian cancer research. He states, “We know that the sensor binds and responds to proteins and small molecules in the blood, but we don’t know which of the proteins or molecules are specific to cancer.” More broadly, the availability of data, or the lack thereof, remains a significant impediment. Ms. Moran observes, “People aren’t sharing their data, or there’s not a mechanism to do it.” Ocra is financing a comprehensive patient registry, incorporating electronic medical records of patients who have consented to allow researchers to train algorithms using their data. Ms. Moran concludes, “It’s early days – we’re still in the wild west of AI now.” Copyright 2024 BBC. All rights reserved. The BBC is not responsible for the content of external sites. Read about our approach to external linking.