Lung cancer is a fatal condition. Lung cancer is one of the leading causes of mortality worldwide, accounting for approximately 2.21 million cases in 2020 alone, according to the World Health Organization. It’s also important to note that the condition can progress, which means that for many people, it may first only manifest as minor symptoms that go unnoticed before rapidly developing into a fatal diagnosis. Fortunately, over the past 20 years, there has been a significant expansion in the range of therapies available to individuals with lung cancer. However, one of the only ways to drastically lower mortality rates is still through early cancer identification.
The development of a deep learning model called “Sybil” by the Massachusetts Institute of Technology (MIT) and Mass General Hospital (MGH) that may be used to predict lung cancer risk using data from just one CT scan is one major achievement in this field. The research, which was last week formally published in the Journal of Clinical Oncology, describes how “tools that give tailored future cancer risk assessment should direct efforts toward those most likely to benefit.” So, according to the study’s authors, “a deep learning model assessing the complete volumetric LDCT [Low Dose Contrast CT] data might be created to predict individual risk without requiring extra demographic or clinical data.”
The model’s basic concept is that “LDCT images contain information that is predictive of future lung cancer risk beyond currently identifiable signs such as lung nodules.” Because of this, the researchers set out to “create and evaluate a deep learning system that predicts future lung cancer risk out to 6 years from a single LDCT scan, and assess its potential therapeutic benefit.”
Overall, the project has been surprisingly effective so far, with Sybil being able to accurately forecast a patient’s future risk of developing lung cancer with just one LDCT’s worth of data.
Without a doubt, this technology still has undeveloped clinical uses and ramifications. Even the study’s authors concur that much effort will be required to determine the precise application of this technology in clinical practice— particularly in terms of building the technology’s credibility so that doctors and patients would feel secure relying on the system’s outputs.
The algorithm’s underlying idea is still highly potent and could revolutionize the field of predictive diagnostics.
Never before have diagnostic tools been so effective. The ability to forecast a long-term disease function using just one CT scan could potentially solve a variety of issues, the most crucial of which is enabling early treatment and lowering mortality.
Pundits may first criticize these types of systems, stating that no artificial intelligence (AI) system could possibly equal the judgment and clinical acumen of a human doctor in a way that would allow them to be replaced. But the goal of these kinds of technologies isn’t necessarily to replace medical knowledge, but rather to maybe improve medical workflows.
A system like Sybil may easily be utilized as a suggestion tool, notifying doctors to potentially alarming CT scans and letting them decide whether or not to follow Sybil’s recommendation based on their own clinical judgment. This could serve as a secondary “check” procedure, which could potentially increase diagnosis accuracy, in addition to most certainly improving clinical throughput.
There is clearly still a lot of work to be done in this area. In addition to improving the algorithm and system itself, scientists, developers, and innovators have a long road ahead of them as they attempt to bring this technology into use in real clinical settings. However, if the technology is created in a safe, moral, and effective manner, it has the goal and capacity to improve patient care, which bodes well for the next generation of diagnostics.
