Lung cancer is the leading cause of cancer-related deaths worldwide, however, advancements in technology are offering new found hope. Over the next decades, AI is poised to revolutionize lung cancer outcomes by detecting cancer at its earliest stages. By improving earlier detection and accelerating identification of the high-risk patient, this technology has the capability to dramatically lower mortality rates.
The National Lung Screening Trial (NLST), published in 2011, showed that a 20% reduction in Lung Cancer Deaths was possible when high risk populations were screened using CT. This discovery is what pushed the National Preventative Services Task Force to recommend Lung Cancer screening high risk patients, which in turn triggered reimbursement for CT based Lung Cancer screening by Medicare. It was at this same time, the first AI system for detection of lung nodules in CT scans was approved by the FDA, marking the beginning of the connection between AI and lung cancer diagnosis.
Today, over a dozen nodule detectors are now being sold, in the US or Europe. CT resolution has improved, radiation dose has decreased, and the early detection of lung cancer through CT screening has been shown to save lives. A 20-year follow up study published by I-ELCAP showing that Lung Cancer detected early by CT Screening can essentially be cured. The reported 20-year survival was just over 80%. This is a huge leap from the traditional methods, which are often impacted by late-stage diagnosis.
We now have a highly sensitive, non-invasive test in CT screening, which is less than $150 for Medicare Patients, and available nearly everywhere in developed nations. AI has further improved sensitivity, while simultaneously making radiologists faster.
So what’s next? While we’ve made significant progress, two key challenges remain. First, while sensitivity is high for the detection of Lung Cancer, specificity is quite low. Second, identification of high risk patients for screening can be quite challenging. Fortunately, new data is showing that AI can help address both of these challenges.
The Low Specificity Problem
Only about 1% of small nodules detected on CT are cancer, and all of these false positives create a huge problem for health systems.
Ruling out cancer by dramatically increasing lung biopsies is not the answer, as lung biopsies are quite invasive and expensive, so when nodules are found (and they are found in about 40% of Chest CT Scans), each nodule is tracked.
To track a patient’s nodule, a follow up scan is performed months later (frequently multiple times) in an attempt to detect cancer-like growth. This is expensive and administratively difficult. Complicating things, patient compliance can be quite low.
For the solution, we can look to AI. A study published in 2023 showed that an AI tool can identify high risk and low risk nodules. Smaller nodules, for example, have a less than 1% chance of being malignant. But when flagged by the AI as high risk, the likelihood of malignancy increases to close to 20%.
The Identification of High Risk Patients Problem
While identifying patients at high risk using smoking history makes sense, but data continues to emerge that it is not sufficient. Approximately 20% of lung cancers that are diagnosed this year will be in never-smokers. A 2022 study showed that when incidentally found nodules are tracked, about half of the Lung Cancers found were in patients who did not qualify for Lung Cancer Screening. This was because they were non-smokers or had not smoked long enough to satisfy the screening guidelines.
Lung cancer in never smokers has been rising significantly, as a result, we need to expand our scope of who is considered high-risk to increase early detection. Incorporating multiple risk factors into the decision to screen for lung cancer is something AI can do well by analyzing the vast amounts of data available efficiently. This personalized approach can help us in turning data into diagnosis faster than ever, allowing for expansion of who is able to be identified as a high risk patient.
Lung cancer remains the biggest cancer killer and the prognosis remains quite poor when caught in later stages when patients are symptomatic. We should, however, remain hopeful. CT Scanners can see early stage lung cancer when it can still be cured, and I predict that this year we will see AI being used to solve two of the biggest remaining challenges; identification of non-smoking patients who need screening, and flagging nodules that need immediate attention.
