An artificial intelligence model designed and built by doctors, scientists and researchers is able to accurately identify cancer in a development they say could escalate diagnosis of the disease and fast-track patients to treatment.
Cancer kills about 10m people annually ( nearly one in six deaths), but in many cases, the disease can be cured if detected early and treated swiftly.
The AI tool designed by experts at the Royal Marsden NHS foundation trust, the Institute of Cancer Research, London, and Imperial College London, can identify whether abnormal growths found on CT scans are cancerous, reports The Guardian.
The algorithm performs more efficiently and effectively than current methods, according the findings, published in The Lancet’s eBioMedicine journal.
“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,” said Dr Benjamin Hunter, a clinical oncology registrar at the Royal Marsden and a clinical research fellow at Imperial.
The team used CT scans of about 500 patients with large lung nodules to develop an AI algorithm using radiomics. The technique can extract vital information from medical images not easily spotted by the human eye.
The AI model was then tested to determine if it could accurately identify cancerous nodules.
The study used a measure called area under the curve (AUC) to see how effective the model was at predicting cancer. An AUC of one indicates a perfect model, while 0.5 would be expected if the model was randomly guessing.
The results showed the AI model could 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 0.67. The model also performed comparably with the Herder score – another test – which had an AUC of 0.83.
“According to these initial results, our model appears to identify cancerous large lung nodules accurately,” Hunter said. “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.”
The AI model may also help doctors make quicker decisions about patients with abnormal growths that are currently deemed medium-risk.
When combined with Herder, the AI model was able to identify high-risk patients in this group. It would have suggested early intervention for 18 out of 22 (82%) of the nodules that went on to be confirmed as cancerous, according to the study.
The team stressed that the Libra study – backed by the Royal Marsden Cancer Charity, the National Institute for Health and Care Research, RM Partners and Cancer Research UK – was still at an early stage. More testing will be required before the model can be introduced in healthcare systems.
But its potential benefits were clear, they said. Researchers hope the AI tool would eventually be able to speed up the detection of cancer by helping to fast-track patients to treatment, and by streamlining the analysis of CT scans.
“Through this work, we hope to push boundaries to speed up the detection of the disease using innovative technologies such as AI,” said the Libra study’s chief investigator, Dr Richard Lee.
The consultant physician in respiratory medicine at the Royal Marsden and team leader at the Institute of Cancer Research said lung cancer was a good example of why new initiatives to speed up detection were urgently needed.
Lung cancer is the biggest worldwide cause of cancer mortality, and accounts for a fifth (21%) of cancer deaths in the UK. Those diagnosed early can be treated much more effectively, but recent data show more than 60% of lung cancers in England are diagnosed at either stage three or four.
“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,” said Lee.
“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.”
Study details
A radiomics-based decision support tool improves lung cancer diagnosis in combination with the Herder score in large lung nodules
Benjamin Hunter, Mitchell Chen, Prashanthi Ratnakumar, Esubalew Alemu, Andrew Logan,
Kristofer Linton-Reid, et al.
Published in The Lancet eBioMedicine on 9 November 2022
Summary
Background
Large lung nodules (≥15 mm) have the highest risk of malignancy, and may exhibit important differences in phenotypic or clinical characteristics to their smaller counterparts. Existing risk models do not stratify large nodules well. We aimed to develop and validate an integrated segmentation and classification pipeline, incorporating deep-learning and traditional radiomics, to classify large lung nodules according to cancer risk.
Methods
502 patients from five U.K. centres were recruited to the large-nodule arm of the retrospective LIBRA study between July 2020 and April 2022. 838 CT scans were used for model development, split into training and test sets (70% and 30% respectively). An nnUNet model was trained to automate lung nodule segmentation. A radiomics signature was developed to classify nodules according to malignancy risk. Performance of the radiomics model, termed the large-nodule radiomics predictive vector (LN-RPV), was compared to three radiologists and the Brock and Herder scores.
Findings
499 patients had technically evaluable scans (mean age 69 ± 11, 257 men, 242 women). In the test set of 252 scans, the nnUNet achieved a DICE score of 0.86, and the LN-RPV achieved an AUC of 0.83 (95% CI 0.77–0.88) for malignancy classification. Performance was higher than the median radiologist (AUC 0.75 [95% CI 0.70–0.81], DeLong p = 0.03). LN-RPV was robust to auto-segmentation (ICC 0.94). For baseline solid nodules in the test set (117 patients), LN-RPV had an AUC of 0.87 (95% CI 0.80–0.93) compared to 0.67 (95% CI 0.55–0.76, DeLong p = 0.002) for the Brock score and 0.83 (95% CI 0.75–0.90, DeLong p = 0.4) for the Herder score. In the international external test set (n = 151), LN-RPV maintained an AUC of 0.75 (95% CI 0.63–0.85). 18 out of 22 (82%) malignant nodules in the Herder 10–70% category in the test set were identified as high risk by the decision-support tool, and may have been referred for earlier intervention.
Interpretation
The model accurately segments and classifies large lung nodules, and may improve upon existing clinical models.
The Guardian article – New artificial intelligence tool can accurately identify cancer (Open access)
See more from MedicalBrief archives:
Late diagnosis by GPs link to lung cancer mortality — UK report
Cancer screening – the good, the bad and the ugly
Cancer: Survivability is changing fast