The Growing Power of Machine Learning: Combating CNS Tumors
Written by Haylie Wilcox
Edited by Khushi Shah
April 5, 2024
Edited by Khushi Shah
April 5, 2024
Research
Inside the operating room, a neurosurgeon confronts one of the most challenging moments of their career. Their patient lies unconscious on the surgical table, concealing a central nervous system (CNS) tumor within their brain. These tumors pose a significant threat, especially to children, making them one of the deadliest forms of cancer (Vermeulen et al., 2023). The lack of immediate information has always been a surgeon's opponent during such procedures, leaving them in the dark about the nature of the tumor. Every moment is critical, but the traditional postoperative pathology reports can’t provide real-time insights. The neurosurgeon needs an ally, and in that room, cutting-edge technology revolutionizes their approach.
This is the story of how machine learning is transforming CNS tumor surgery, providing real-time data and guidance to surgeons in their pursuit to conquer the enemy lurking within the human brain.
CNS tumors are tumors in the brain and spinal cord, ranging from benign to malignant. These tumors affect vital functions and are often life-threatening. Symptoms can be diverse, including changes in vision, persistent headaches, seizures, and weakness or numbness in the arms or legs. Prompt and precise diagnosis is essential, as these symptoms may mimic other conditions. Current diagnosis relies on preoperative (before the operation) imaging, such as MRI or CT scans, and intraoperative (during the operation) diagnosis including the assessment of frozen tumor sections on a cellular level. However, this method has limitations, particularly during surgery when quick decisions are crucial. Often without a clear diagnosis postoperative pathology is needed to revise the diagnosis, leading to a second surgery (Vermeulen et al., 2023) that may increase patient risks and complexities in the treatment process.
Enter "Sturgeon," the machine learning model designed for intraoperative CNS tumor classification (Vermeulen et al., 2023). DNA methylation, an epigenetic modification of DNA molecules, plays a pivotal role in Sturgeon’s design. DNA methylation occurs when methyl groups are added to the DNA molecule creating patterns in cancer cells that can be distinct from normal cells, serving as unique fingerprints for tumor types. These patterns can be decoded to categorize tumors with high accuracy. Sturgeon also uses Nanopore sequencing, an innovative technology that rapidly reads DNA and offers real-time results, making it invaluable in clinical settings. This technology is cost-effective and provides valuable insights to act as an aid. Sturgeon leverages machine learning to analyze DNA methylation patterns via nanopore sequencing in real-time during surgery, making precise classification decisions accessible to neurosurgeons (Vermeulen et al., 2023). The study showcases Sturgeon's real-time performance during 25 surgeries, with a track record of 18 accurate diagnoses in less than 90 minutes (Vermeulen et al., 2023). However, in 7 cases, Sturgeon didn't achieve the necessary confidence threshold for a conclusive diagnosis (Vermeulen et al., 2023). Although Sturgeon didn't reach the required confidence threshold in some cases, it seems to remain a valuable and evolving tool that enhances surgical decision-making. This tool’s ability to ensure high accuracy within minutes solidifies its role as a vital asset in the fight against CNS tumors. Beyond CNS tumors, the applications of this technology may extend to other areas of medicine. The ability to rapidly diagnose and classify tumors may open the door for faster treatment initiation and better patient care.
While machine learning in surgery offers immense potential, it also comes with challenges. Ethical considerations play a crucial role in ensuring the responsible use of this technology, as they can provide a framework that safeguards against potential bias and promotes transparency. The power of machine learning is rewriting the story of CNS tumor surgery. It reduces the risks of misdiagnosis, guiding surgeons with more information when faced with ambiguous cases. As technology continues to advance, embracing these tools remains essential to transform healthcare and improve patient outcomes.
In this age of rapid technological progress, staying informed about machine learning and AI’s role in healthcare is crucial. Embracing these innovations can lead to better patient care, ultimately saving lives and ensuring a healthier future.
References
“Brain and Spinal Cord Tumors.” National Institute of Neurological Disorders and Stroke, U.S. Department of Health and Human Services, www.ninds.nih.gov/health-information/disorders/brain-and-spinal-cord-tumors.
Jin, B., Li, Y., & Robertson, K.D. (2011). DNA methylation: superior or subordinate in the epigenetic hierarchy? Genes Cancer. 2 (6): 607-17. doi: 10.1177/1947601910393957. PMID: 21941617; PMCID: PMC3174260.
PDQ® Adult Treatment Editorial Board. PDQ Adult Central Nervous System Tumors Treatment. Bethesda, MD: National Cancer Institute. Updated 09/07/2023. Available at: https://www.cancer.gov/types/brain/patient/adult-brain-treatment-pdq.
Van der Reis, A. L., Beckley, L. E., Olivar, M. P., & Jeffs, A. G. (2023). Nanopore short-read sequencing: A quick, cost-effective, and accurate method for DNA metabarcoding. Environmental DNA, 5, 282–296. https://doi.org/10.1002/edn3.374
Vermeulen, C., Pagès-Gallego, M., Kester, L. et al. (2023). Ultra-fast deep-learned CNS tumor classification during surgery. Nature 622, 842–849. https://doi.org/10.1038/s41586-023-06615-2
This is the story of how machine learning is transforming CNS tumor surgery, providing real-time data and guidance to surgeons in their pursuit to conquer the enemy lurking within the human brain.
CNS tumors are tumors in the brain and spinal cord, ranging from benign to malignant. These tumors affect vital functions and are often life-threatening. Symptoms can be diverse, including changes in vision, persistent headaches, seizures, and weakness or numbness in the arms or legs. Prompt and precise diagnosis is essential, as these symptoms may mimic other conditions. Current diagnosis relies on preoperative (before the operation) imaging, such as MRI or CT scans, and intraoperative (during the operation) diagnosis including the assessment of frozen tumor sections on a cellular level. However, this method has limitations, particularly during surgery when quick decisions are crucial. Often without a clear diagnosis postoperative pathology is needed to revise the diagnosis, leading to a second surgery (Vermeulen et al., 2023) that may increase patient risks and complexities in the treatment process.
Enter "Sturgeon," the machine learning model designed for intraoperative CNS tumor classification (Vermeulen et al., 2023). DNA methylation, an epigenetic modification of DNA molecules, plays a pivotal role in Sturgeon’s design. DNA methylation occurs when methyl groups are added to the DNA molecule creating patterns in cancer cells that can be distinct from normal cells, serving as unique fingerprints for tumor types. These patterns can be decoded to categorize tumors with high accuracy. Sturgeon also uses Nanopore sequencing, an innovative technology that rapidly reads DNA and offers real-time results, making it invaluable in clinical settings. This technology is cost-effective and provides valuable insights to act as an aid. Sturgeon leverages machine learning to analyze DNA methylation patterns via nanopore sequencing in real-time during surgery, making precise classification decisions accessible to neurosurgeons (Vermeulen et al., 2023). The study showcases Sturgeon's real-time performance during 25 surgeries, with a track record of 18 accurate diagnoses in less than 90 minutes (Vermeulen et al., 2023). However, in 7 cases, Sturgeon didn't achieve the necessary confidence threshold for a conclusive diagnosis (Vermeulen et al., 2023). Although Sturgeon didn't reach the required confidence threshold in some cases, it seems to remain a valuable and evolving tool that enhances surgical decision-making. This tool’s ability to ensure high accuracy within minutes solidifies its role as a vital asset in the fight against CNS tumors. Beyond CNS tumors, the applications of this technology may extend to other areas of medicine. The ability to rapidly diagnose and classify tumors may open the door for faster treatment initiation and better patient care.
While machine learning in surgery offers immense potential, it also comes with challenges. Ethical considerations play a crucial role in ensuring the responsible use of this technology, as they can provide a framework that safeguards against potential bias and promotes transparency. The power of machine learning is rewriting the story of CNS tumor surgery. It reduces the risks of misdiagnosis, guiding surgeons with more information when faced with ambiguous cases. As technology continues to advance, embracing these tools remains essential to transform healthcare and improve patient outcomes.
In this age of rapid technological progress, staying informed about machine learning and AI’s role in healthcare is crucial. Embracing these innovations can lead to better patient care, ultimately saving lives and ensuring a healthier future.
References
“Brain and Spinal Cord Tumors.” National Institute of Neurological Disorders and Stroke, U.S. Department of Health and Human Services, www.ninds.nih.gov/health-information/disorders/brain-and-spinal-cord-tumors.
Jin, B., Li, Y., & Robertson, K.D. (2011). DNA methylation: superior or subordinate in the epigenetic hierarchy? Genes Cancer. 2 (6): 607-17. doi: 10.1177/1947601910393957. PMID: 21941617; PMCID: PMC3174260.
PDQ® Adult Treatment Editorial Board. PDQ Adult Central Nervous System Tumors Treatment. Bethesda, MD: National Cancer Institute. Updated 09/07/2023. Available at: https://www.cancer.gov/types/brain/patient/adult-brain-treatment-pdq.
Van der Reis, A. L., Beckley, L. E., Olivar, M. P., & Jeffs, A. G. (2023). Nanopore short-read sequencing: A quick, cost-effective, and accurate method for DNA metabarcoding. Environmental DNA, 5, 282–296. https://doi.org/10.1002/edn3.374
Vermeulen, C., Pagès-Gallego, M., Kester, L. et al. (2023). Ultra-fast deep-learned CNS tumor classification during surgery. Nature 622, 842–849. https://doi.org/10.1038/s41586-023-06615-2