人工智慧經過訓練可以發現血液測試中的警訊

Artificial intelligence, commonly known as AI, is a powerful tool that has the potential to revolutionize the way doctors analyze blood tests. This is particularly significant in the context of diseases like ovarian cancer, which is often difficult to detect in its early stages. Audra Moran, the leader of the Ovarian Cancer Research Alliance, emphasizes the importance of early detection, stating that the earlier ovarian cancer is identified, the better the chances of successful treatment. Most cases of ovarian cancer begin in the fallopian tubes, and by the time it reaches the ovaries, it may have already spread to other areas of the body. Ms. Moran points out that ideally, ovarian cancer should be detected five years before any symptoms appear in order to significantly improve survival rates. Fortunately, advancements in blood testing technology are emerging, utilizing the capabilities of AI to identify early signs of ovarian cancer. Additionally, AI is also being employed to enhance the speed and accuracy of tests for other serious infections, such as pneumonia. Dr. Daniel Heller, a biomedical engineer at Memorial Sloan Kettering Cancer Center in New York, is at the forefront of this innovative research. His team is developing a groundbreaking testing technology that incorporates nanotubes—extremely tiny tubes made of carbon that are approximately 50,000 times smaller than a human hair. The discovery of these nanotubes dates back about 20 years, when scientists found that they could emit fluorescent light. In recent years, researchers have learned how to modify the properties of these nanotubes so that they can respond to various substances present in blood samples. This allows for the introduction of millions of nanotubes into a blood sample, which can then emit different wavelengths of light depending on what binds to them. However, interpreting the signals generated by these nanotubes poses a challenge, as Dr. Heller likens it to matching a fingerprint. In this scenario, the fingerprint represents a unique pattern of molecules that attach to the sensors, each with varying sensitivities and binding strengths. Unfortunately, these patterns are often too subtle for human observers to detect. Dr. Heller explains, 'We can look at the data and we will not make sense of it at all. We can only see the patterns that are different with AI. ' To decode the data from the nanotubes, the team employs a machine-learning algorithm, training it to distinguish between samples from patients with ovarian cancer and those from individuals without the disease. This training set includes blood samples from patients with other types of cancer or gynecological conditions that could be mistaken for ovarian cancer. One of the significant challenges in utilizing AI for developing blood tests for ovarian cancer is the rarity of the disease, which limits the amount of data available for training the algorithms. Furthermore, much of the existing data is often confined within hospitals, with minimal sharing among researchers. Dr. Heller describes the process of training the algorithm on data from just a few hundred patients as a 'Hail Mary pass,' indicating the high level of uncertainty involved. Nevertheless, he reports that the AI achieved greater accuracy than the best cancer biomarkers currently available, and this was only the initial attempt. The system is undergoing further studies to determine if it can be enhanced by incorporating larger sets of sensors and samples from a broader range of patients. Just as algorithms for self-driving cars improve with more real-world testing, more data can enhance the performance of the AI. Dr. Heller is optimistic about the future of this technology, stating, 'What we'd like to do is triage all gynecological 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. ' He anticipates that this capability may be realized within the next three to five years. Beyond early detection, AI is also proving to be valuable in expediting other blood tests. For cancer patients, the risk of pneumonia can be life-threatening, and with around 600 different organisms capable of causing pneumonia, doctors often need to conduct multiple tests to identify the specific infection. However, new blood testing methods are simplifying and accelerating this process. Karius, a company based in California, utilizes AI to accurately identify the specific pneumonia pathogen within 24 hours, allowing for the selection of the appropriate antibiotic. Alec Ford, the CEO of Karius, explains, 'Before our test, a patient with pneumonia would have 15 to 20 different tests to identify their infection in just their first week in the hospital—that's about $20,000 in testing. ' Karius has developed a vast database of microbial DNA containing tens of billions of data points, enabling lab workers to compare patient samples against this database to pinpoint the exact pathogen. Mr. Ford asserts that this level of precision would not have been achievable without the assistance of AI. However, researchers face challenges in fully understanding the connections that AI may establish between test biomarkers and diseases. Over the past two years, Dr. Slavé Petrovski has created an AI platform called Milton, which utilizes biomarkers from the UK Biobank data to identify 120 diseases with a remarkable success rate exceeding 90%. The ability to discern patterns within such extensive data is a task that only AI can accomplish. Dr. Petrovski, a researcher at 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 employs a similar pattern-matching technique in his research on ovarian cancer, acknowledging that while the sensors respond to proteins and small molecules in the blood, the specific proteins or molecules linked to cancer remain unknown. More broadly, the issue of data availability continues to be a significant hurdle. Ms. Moran highlights that many individuals are reluctant to share their data, and there is often no established mechanism for doing so. To address this, Ocra is funding a large-scale patient registry that will compile electronic medical records of patients who have consented to allow researchers to utilize their data for algorithm training. Ms. Moran concludes, 'It's early days—we're still in the wild west of AI now.