The most significant cause of cancer-related death worldwide is lung cancer. Globally, 13 per cent of all new cancer cases and 19 per cent of all cancer-related fatalities are attributable to it. In addition, an estimated 1.8 million people were diagnosed with lung cancer for the first time in 2012. It is the most common malignancy and the primary cause of cancer-related mortality in men in India, with the highest incidence in Mizoram in both sexes (age-adjusted rate 28.3 and 28.7 per 100,000 population in males and females, respectively).
Lung cancer incidence has increased in both sexes in Delhi, Chennai, and Bengaluru. To a considerable extent, the frequency and pattern of smoking account for the variation in lung cancer incidence and pattern among geographic regions and ethnic groups.
Lung cancer has a terrible five-year survival rate, hovering at 15 per cent in affluent countries and five per cent in developing ones. Low-dose computed tomography (CT) reduces lung cancer mortality by 20 per cent in high-risk populations. However, this benefit comes at the expense of a 96 per cent false-positive rate. The usefulness of such a screening tool in a country like India, where tuberculosis is rampant, seems dubious. There is a need to discover novel non-invasive methods/ biomarkers for the early diagnosis and screening of the high-risk group.
Recently, scientists developed the Sybil deep-learning cancer risk model. In contrast to existing methods, Sybil only takes a single low-chest computed tomography (CT) scan to predict lung cancer risk one to six years after screening.
Dr Lecia V Sequist, professor of medicine at Harvard Medical School, stated that: "Sybil provides a risk score, not a diagnosis, so it's most useful for identifying which patients need to be followed closely or screened for cancer."
Along with image annotations, Sybil's method is made available to the public to encourage further study and therapeutic applications.
How do they predict it?
Using data from 15,000 users, the researchers created an AI model with deep learning. In total, 35,001 low-dosage CT images were utilised for training and constructing their model, and 6,282 were used to test it.
To train the model, two thoracic radiologists marked worrisome lesions on scans of patients who got cancer within a year after the scan.
Sybil had a correct allocation of lung cancer or not of 92 per cent across all test data sets after one year, 86 per cent after two years, and a likelihood (C-index) of 75 per cent after six years using only low-dose CT scans.
The researchers observed that Sybil's performance was consistent regardless of her gender, age, or smoking history. Next, they evaluated Sybil using data from the Massachusetts General Hospital (MGH) in Boston, Massachusetts, and the Chang Gung Memorial Hospital (CGMH) in Taiwan. In contrast to the main and MGH datasets, CGMH patients were not required to have a smoking history to undergo a low-dose CT scan.
Sybil accurately identified 86 per cent of lung cancer cases or healthy lungs within a year based on the MGH dataset and 94 per cent of patients based on the CGMH dataset. After six years, it also predicted 81 per cent of lung tumours or healthy lungs in the MGH cohort and 80 per cent in the CGHM cohort. According to the researchers, Sybil may also anticipate typical clinical risk factors such as smoking based on scans.
The researchers acknowledged their model's flaws. They highlighted, for instance, that 92 per cent of Sybil's training data came from White patients, implying that their findings may not apply to populations of other races. They also noted that the training data scans were collected between 2002 and 2004, indicating that advancements in CT technology may harm Sybil's prediction capacity.
Conclusions on Sybil's capacity to predict lung cancer in non-smokers are speculative due to the lack of specific smoking data from CGMH patients. In addition, Sybil could operate in the background at radiology reading stations and forecast lung cancer risk as soon as low-dose CT scans are accessible without the presence of radiologists to annotate regions of interest or demographic and other clinical data, according to the researchers.
The researchers believe Sybil can reduce the need for follow-up scans or biopsies in low-risk patients. Furthermore, they will initiate a prospective clinical trial to evaluate Sybil in the real world and determine how it complements radiologists' jobs. The code has also been made available to the public.
Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!
