Abnormal Laboratory Results

AI Improves Lung Nodule Detection On Chest X-rays (IMAGE)

Caption

Images in a 60-year-old woman who underwent chest radiography for health checkup purposes and was allocated to the artificial intelligence (AI) group. (A) Frontal chest radiograph shows a subtle nodular opacity (arrow) in the right middle lung zone. (B) The lesion was detected by the AI-based computer-aided detection software, with an abnormality probability of 81.1%. The designated radiologist reported this chest radiograph as positive. (C) Axial, noncontrast, low-dose chest CT scan shows a 1.1-cm solid nodule (arrow) in the right lower lobe. The patient underwent percutaneous needle biopsy, and the nodule was confirmed to be adenocarcinoma.

Disclaimer: AAAS and EurekAlert! Are not responsible for the accuracy of news releases posted to EurekAlert! By contributing institutions or for the use of any information through the EurekAlert system.

AI Could Help Doctors Diagnose Lung Cancer

Artificial intelligence (AI) could help doctors diagnose lung cancer earlier, according to a study led by researchers from The Royal Marsden NHS Foundation Trust in collaboration with The Institute of Cancer Research, London, and Imperial College London.

The LIBRA study – which was supported by The Royal Marsden Cancer Charity, the National Institute for Health and Care Research (NIHR), RM Partners and Cancer Research UK – used data from the CT scans of nearly 500 patients with large lung nodules to develop an AI algorithm. The AI model was then tested to see if it could accurately identify cancerous nodules. 

Lung nodules are abnormal growths that are common and mostly benign. However, some lung nodules can be cancerous, and large ones (e.G. 15-30mm in size) are associated with the highest risk.

Speed up detection of lung cancer

Researchers hope this technology will eventually be able to speed up the detection of lung cancer by helping to fast-track high-risk patients to treatment, and by streamlining the analysis of patient scans. 

The authors used a measure called the AUC ("Area under the curve") to assess how effective the new model was at predicting cancer. An AUC of 1 indicates a perfect model, while 0.5 would be expected if the model was randomly guessing. The results, which have been published in the Lancet's eBioMedicine, indicate that the AI model was able to identify each nodule's risk of cancer with an AUC of 0.87. The performance improved on the Brock score, a test currently used in clinic, which scored 0.672. 

The new model also performed comparably to the Herder score, another test currently used in clinic, which had an AUC of 0.83. However, as the artificial intelligence model uses only two variables, as opposed to 7 for the Herder score and 9 for the Brock score, it could potentially streamline and speed up nodule risk calculation in the future. 

The new model may also help clinicians make decisions about patients that currently don't have a clear referral pathway. Using Herder, patients are categorised as low risk if they score less than 10 per cent, and high risk and needing intervention if they score over 70 per cent. For the patients in the intermediate risk group (10-70 per cent), a broad range of tests or treatment options could be considered. When combined with Herder, the researchers' model was able to identify high-risk patients in this group and would have suggested early invention for 18 out of 22 (82 per cent) of the nodules that went on to be diagnosed as cancerous. 

Extracting data from medical images

To analyse the CT scan data, the researchers used a technique called radiomics, which can extract information about the patient's disease from medical images that can't be easily seen by the human eye. 

Lung cancer is the leading worldwide cause of cancer mortality and accounts for just over a fifth (21 per cent) of cancer deaths in the UK. Patients diagnosed with early-stage disease can be treated much more effectively, but recent data shows over 60 per cent of lung cancers in England are diagnosed at either stage three or four, so initiatives to speed up detection are urgently needed. 

Dr Benjamin Hunter, Clinical Oncology Registrar at The Royal Marsden NHS Foundation Trust and Clinical Research Fellow at Imperial College London, who is funded by Cancer Research UK, said:

"According to these initial results, our model appears to identify cancerous large lung nodules accurately. In the future, we hope it will improve early detection and potentially make cancer treatment more successful by highlighting high-risk patients and fast-tracking them to earlier intervention. Next, we plan to test the technology on patients with large lung nodules in clinic to see if it can accurately predict their risk of lung cancer."

Innovative AI technologies

Chief investigator for the LIBRA study, Dr Richard Lee, Consultant Physician in Respiratory Medicine and Early Diagnosis at The Royal Marsden NHS Foundation Trust and Team Leader for the Early Diagnosis and Detection team at The Institute of Cancer Research, London, who is funded by The Royal Marsden Cancer Charity, said:

"While at an early stage, this study is an example of the vital scientific clinical research we're undertaking in the Early Diagnosis and Detection Centre at The Royal Marsden and the ICR. Through this work, we hope to push boundaries to speed up the detection of the disease using innovative technologies such as AI.  

"People diagnosed with lung cancer at the earliest stage are much more likely to survive for five years, when compared with those whose cancer is caught late. This means it is a priority we find ways to speed up the detection of the disease, and this study – which is the first to develop a radiomics model specifically focused on large lung nodules – could one day support clinicians in identifying high-risk patients."

Keith Hewett, 64 from Watford, was diagnosed with lung cancer in 2018 and treated with surgery at his local hospital. He was then referred to Dr Richard Lee at The Royal Marsden for follow-up care. Last year, a CT scan revealed nodules in Keith's lung, and, after further investigation, he was diagnosed with cancer again. Keith, whose medical history is similar to patients used in this study, said: 

"After my first diagnosis, I had a CT scan at The Royal Marsden every three months and, just as they were about to become every six months, Dr Lee noticed a change on the scans. They weren't sure what it was but agreed it needed further investigation. As I have arthritis, I can get lumps on my body which added more cloudiness to what was happening.  "It turned out that there were three nodules in my lungs which were cancerous, and I was treated with surgery at The Royal Brompton. My care at The Royal Marsden has been excellent as their attention to detail is great and I felt safe in their care. 

"Any new technology that helps gives more clarity over whether something on a CT scan is or isn't cancer would be great. As a patient, you want to know whether you have the disease as soon as possible because the earlier the treatment, the better the outcome."

Genes, Smoking, And Lung Cancer

Nature Publishing Group Video: Genetic Link Between Smoking and Lung Cancer

So, what specific genetic factors are involved in an increased risk of lung cancer? In recent years, three independent groups of international scientists have identified a region on chromosome 15 that, if mutated, dramatically increases a smoker's risk of developing lung cancer by another 30% to 80% (giving smokers who carry this mutation an overall lung cancer risk of about 20% to 23%), depending on whether an individual has one or two copies of what the researchers are calling the 15q24 susceptibility locus (Amos et al., 2008; Hung et al., 2008; Thorgeirsson, et al. 2008). In general, a susceptibility locus is a region on a given chromosome where mutations that affect one or more genes are suspected to be present, based on statistical evidence. These mutations can be located in coding segments of one or more genes, therefore directly affecting gene products, or they may be found in sequences that control gene function (known as regulatory regions).

Altogether, the three research groups that studied the 15q24 susceptibility locus surveyed more than 35,000 people across Europe, the United States, and Canada. Data were collected on both lung cancer patients and people without lung cancer, as well as on both smokers and nonsmokers. All three groups of researchers eventually zeroed in on the same section of DNA: the long arm of chromosome 15, a region that encodes several genes, including a few that code for nicotinic acetylcholine receptors. These receptors bind to nicotine and nicotine derivatives and are found on cells in the nervous system, in the lungs, and elsewhere in the body. Moreover, all three groups of researchers used the same basic technique in isolating this portion of chromosome 15: genome-wide association studies (GWAS).

Scientists have used GWAS to discover over 100 regions of the genome now known to be associated with various fairly complex human disorders (i.E., complex with respect to the number of genes involved and the amount of interaction between those genes and the environment), including diabetes, heart disease, and breast, colorectal, and prostate cancers. In the three lung cancer studies, the researchers used GWAS to scan the genome, and they found associations between single nucleotide polymorphisms (SNPs) and lung cancer on chromosome 15. In other words, the researchers found that certain SNPs are more common in people with lung cancer. (SNPs are DNA sequence variations that result from one nucleotide base being substituted by another; scientists estimate that as much as 90% of all human genetic variation is in the form of SNPs.)

In one of the three studies, the scientists conducted genome-wide SNP genotyping on 14,000 Icelandic smokers and then evaluated associations between SNP variation and lung cancer, as well as between SNP variation and the number of cigarettes smoked per day. These studies yielded two interesting findings. First, the researchers noted a significant association between variation at the 15q24 locus and lung cancer; second, they found a significant correlation between that same variation and the average number of cigarettes an individual smoked each day. This second finding led the researchers to argue that there is genetic basis for both lung cancer and nicotine dependence. In other words, based on the researchers' interpretation of the evidence, not only do genes ("nature") play a causal role in lung cancer, alongside the huge causal role that smoking ("nurture") has long been known to occupy, but they also seem to have something to do with why people become addicted to tobacco in the first place.

The other two groups of scientists agreed with some of the first group's findings; specifically, they also concluded that SNP variation is correlated with lung cancer risk and that smokers with a particular type of SNP genotype have a substantially greater risk of developing lung cancer than individuals with other genotypes. However, these two sets of scientists came to a much different conclusion than the first group regarding the genetic basis for tobacco addiction. While these scientists did not actually investigate the specific roles of any of the genes in the 15q24 region, they argued that the evidence provided sufficient reason to speculate that the altered genes on 15q24 most likely play a direct causal role in lung cancer by interfering with nicotine acetylcholine receptors and stimulating tumor growth (as opposed to an indirect causal role related to nicotine addiction). For example, in the study from one laboratory, which involved about 12,000 individuals from several central European countries, the scientists found a weak but still statistically significant association between variation in 15q24 and lung cancer in nonsmokers (Hung et al., 2008). The authors speculated that the presence of this association in both nonsmokers and smokers suggests that the association between the gene and lung cancer is direct and has little to do with nicotine addiction. If the association were mediated through nicotine addiction, the scientists reasoned, they shouldn't have found the association in nonsmokers. The researchers further argued that a lack of correlation between genes in the 15q24 region and other smoking-related cancers (e.G., head and neck cancer) also suggests that the mechanism is probably not related to nicotine addiction. If it were, they reasoned, there should be an increased risk associated with 15q24 for those other cancers as well.

---