Lung Cancer Warning System: The First of Its Kind

Recently, AI-related headlines have been mainly along the lines of how the technology will steal our jobs and make our lives difficult. And not to play the devil’s advocate, but aren’t we missing the bigger picture here? Artificial intelligence is much more than what mainstream media is having you believe. In fact, it could be the key tool to bolster the coffee industry. Additionally, in more current news, researchers from the Massachusetts Institute of Technology (MIT) and the Massachusetts General Cancer Center trained an AI-based software, dubbed Sybil, to calculate the probability of developing lung cancer within the next six years. So, let’s talk about it!

Understanding Why We Need It

It goes without saying that jobs in the healthcare sector are some of the most mentally and oftentimes physically taxing jobs out there. That said, they are humans, and if there is one thing I know about humans is that we are prone to error. You see, a diagnosis is an educated guess based on over seven years’ worth of experience and accumulated knowledge. In addition, many diseases have common symptoms, and testing is seldom conclusive. A 2019 study determined that 1 in 20 American adults is misdiagnosed yearly. So, it is safe to say that diagnostic medicine is not an exact science.

Lung Cancer Misdiagnosis

According to a study from 2020 that was published in the journal Diagnosis, lung cancer has the highest diagnostic error rates of all cancer types (22.5 percent vs. 2.4 percent for prostate cancer). Lung cancer symptoms include coughing, wheezing, shortness of breath, and a hoarse voice. However, these are also signs of several respiratory conditions, such as pleurisy, bronchitis, pneumonia, asthma, and tuberculosis. In addition, people are not getting routine screenings despite public health initiatives, causing delays in diagnosis. In fact, the U.S. Preventive Services Task Force advises annual scans for people 50 years of age and older who have smoked 20 packs per year in the past, but only a fraction actually abides by this recommendation. So, it is necessary to find ways to increase the rate so that more people can receive timely treatment.

Sybil: The Modern Oracle

Historically, oracles were people or things considered to have provided wise and insightful counsel or prophetic predictions, especially precognition of the future. Today, thanks to machine learning (ML) and deep learning (DP), subsets of Artificial Intelligence, we have our own version of the oracles: Sybil, aptly named after prophetesses or oracles in Ancient Greece.

Before We Continue…

Low-Dose Computed Tomography (LCDT), also called low-dose CT scan, is a procedure that relies on a computer linked to an x-ray machine. This setup gives off a very low dose of radiation, making detailed pictures of areas inside the body. The images are then taken from different angles to create 3D views of tissues and organs. Based on the age and smoking history, it is the only recommended screening for adults with a high risk of developing lung cancer.

What Part Does Sybil Play?

The team designed the AI-based software to address two significant problems with current lung cancer screening guidelines.

• Demographics: The researchers made sure to apply the AI tool to patients with no smoking history. The reason for that is the fact that the guidelines stress the most on screening for current and past smoking despite the fact that the rate of finding lung cancer in nonsmokers’ lungs has increased in the last decade (you might want to blame pollution for that).
• Convenience: But that is not to say that current and former smokers don’t benefit from this. In fact, Sybil will make screening more convenient. As previously stated, the guidelines recommend a yearly scan for them, but only about ten percent do it. Sybil’s ability to predict six years in advance with a high level of accuracy will fix this issue: a single scan every six years. The innovation can help improve the current rates.

How Did They Do It?

Simply put, they fed the deep learning AI algorithm a data set of over 8,8000 LCDTs from Mass General and 18,000 from Chang Gung Memorial Hospital in Taiwan. The later records added a broader range of smoking histories.

Once Sybil was done with the training to recognize the early signs of lung cancer, the researchers tested it through the same set it trained on, plus a new set of over 6,000 never-seen-before scans from the National Lung Screening Trial.

The Results

Area-under-the-curve (AUC), in which a perfect score of 1.0 and random chance can achieve a 0.5, evaluated Sybil’s precision in differentiating between healthy and disease samples. With an average AUC of 0.91 across the data sets, the AI was able to predict cancer within a year, with the mixed Taiwanese data set receiving the highest score. Its average AUC for six-year predictions fell to a still-respectable 0.79, with the algorithm performing better on scans added for the training set than on scans added solely for testing.

The Aftermath

With these encouraging findings from the retrospective study, the researchers announced that they are now preparing to start a prospective clinical trial of the AI. This trial would test Sybil’s effectiveness in real-time as well as see if it can be integrated into radiologists’ current workflows. They’ve released Sybil’s code to the public in the interim.

Final Thoughts

This software is honestly beyond amazing. The world is skeptical about AI’s role in our world. I’m a skeptic myself! But one has to give credit where credit is due. A cure for cancer may be too out of reach right now, but early diagnosis is achievable. The results from the preliminary testing are auspicious! If this AI-based software sticks its landing… Oh, the good it will do! Unfortunately, we might witness a Tug of War between Big Tobacco and Big Pharma with Sybil in the middle. The former will want to suppress it while the latter will promote it, both wanting to feed their bottom line. Fingers crossed that their disagreement won’t overshadow and damage the AI development in diagnostic medicine.

Inside Telecom provides you with an extensive list of content covering all aspects of the Tech industry. Keep an eye on our Medtechsection to stay informed and updated with our daily articles.