December 26, 2022 — Artificial intelligence (AI) could help identify the risk of cancer returning in non-small cell lung cancer (NSCLC) patients using CT scans, according to the latest results from 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 latest phase of the OCTAPUS-AI study used imaging and clinical data from over 900 NSCLC patients from the UK and Netherlands following curative radiotherapy to develop and test machine learning (ML) algorithms to see how accurately the models could predict recurrence1.
The study was supported by the Early Diagnosis and Detection Centre which aims to accelerate early diagnosis of cancer. The Centre has been established at The Royal Marsden in partnership with The Institute of Cancer Research (ICR) and is supported by funding from The Royal Marsden Cancer Charity and the National Institute for Health and Care Research (NIHR)2.
NSCLC makes up nearly five sixths (85%) of lung cancer cases and, when caught early, the disease is often curable3. However, over a third (36%) of NSCLC patients in the UK experience their cancer returning, which is known as recurrence4.
This technology could lead to improved post-treatment follow-up for NSCLC patients based on their risk of recurrence and, eventually, other tumour types too. This may mean recurrence in high-risk patients is identified earlier, leading to faster interventions and improved outcomes, while low-risk patients could be spared unnecessary follow-up scans and hospital visits.
These results, which have been published in npj Precision Oncology, reveal the researchers’ model was better at correctly identifying which NSCLC patients were at a higher risk of recurrence within two years of completing radiotherapy, than a model built on the TNM staging system. TNM, which describes the amount and spread of cancer in a patient's body, is currently the gold-standard in predicting the prognosis of cancer patients.
A measurement known as “area under the curve”(AUC) was used to express the effectiveness of this tool. An AUC of one means the system is right every time; 0.5 is the score you would expect if it was randomly guessing and zero means it is always wrong. This model achieved an AUC of 0.738, improving on the traditional TNM staging technique which scored 0.683.
The imaging data was taken from treatment planning CT scans, which all NSCLC patients have prior to radiotherapy5. To analyse this data, researchers used a technique called radiomics, which can extract prognostic information about the patient’s disease from medical images that can’t be seen by the human eye. Data from this technique can also potentially be linked with biological markers6. As a result, researchers believe radiomics could be a useful tool in both personalising medicine as well as improving post-treatment surveillance.
Study lead Dr Sumeet Hindocha, Clinical Oncology Specialist Registrar at The Royal Marsden NHS Foundation Trust, and Clinical Research Fellow at Imperial College London, said:
“While at a very early stage, this work suggests that our model could be better at correctly predicting tumour regrowth than traditional methods. This means that, using our technology, clinicians may eventually be able to identify recurrence earlier in high-risk patients.
“Next, we want to explore more advanced machine learning techniques, such as deep learning, to see if we can get even better results. We then want to test this model on newly diagnosed NSCLC patients and follow them to see if the model can accurately predict their risk of recurrence.”
Dr Sumeet Hindocha is funded by the UKRI Artificial Intelligence for Healthcare Centre for Doctoral Training at Imperial College London.
Chief investigator for the OCTAPUS-AI 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:
“AI is an exciting new frontier in cancer research and The Royal Marsden is leading the way in utilising this technology to diagnose, detect, and prevent the disease more effectively.
“This innovative study demonstrates how AI could advance our ability to predict cancer’s behaviour, including whether an individual is likely to relapse, potentially helping NSCLC patients live longer, and reduce the impact the disease has on their lives.
“In the future, we hope this approach will pave the way for predicting recurrence for all cancer types, not just NSCLC. Our model used features specific to this disease but by refining the algorithm, this technology could have much wider application.”
Retired teacher Les Bennett, 69 from London, is a Royal Marsden patient who has been diagnosed with non-small cell lung cancer three times. Her treatment has included surgery, radiotherapy and briefly Nivolumab, a type of immunotherapy. Les, whose medical history matches the profile of patients in this study, said:
“Back in 2003, I was diagnosed with hairy cell leukaemia, a very rare type of cancer which is not curable but highly treatable. During a follow-up scan in 2014, doctors at my local hospital spotted something unusual on my lungs known as ground-glass opacities. These are usually benign but, as I later found out, can be a sign of cancer. This was devastating to hear as my father died of lung cancer when I was ten. It felt like a potential death sentence. I continued to be closely monitored and, in 2016, I was actually diagnosed with cancer in my left lung. The second time was in 2018, which affected my right lung, and the third time was in 2020, which is when I was referred to The Royal Marsden. I was treated with radiotherapy and nivolumab through the STILE trial, which I stopped after four months as it caused some inflammation.
“The monitoring of my disease including at The Royal Marsden has been excellent and I think having a tailored approach is crucial. I’ve been scanned around every three months since 2016 and, as I’m presently stable, I will be checked again in six months. Being regularly checked has ensured that, each time I’ve been diagnosed, I’ve been able to start treatment promptly. This recent extension in follow-up time is really helpful as I feel sheer terror in the lead up to my scan results and usually experience extreme anxiety. The Royal Marsden are aware of my ‘scanitis’ and have been very understanding and supportive about it. For example, they have referred me for psychological support for this and ensure I receive results as quickly and sensitively as possible.”
These results follow data from the study published earlier this year which used clinical data alone to predict recurrence in this patient group7.
For more information: https://www.royalmarsden.nhs.uk/
1. ML is a type of AI that allows software to automatically predict outcomes. ML algorithms build a model based on sample data to make predictions or decisions without being explicitly programmed to do so.
2. The Early Diagnosis and Detection Centre brings together early detection research and expertise across multiple tumour groups, with Royal Marden Cancer Charity funding supporting the recruitment of new specialist roles and the setup of a new clinical trials infrastructure. It focuses on identifying higher risk groups, who will then benefit from AI, imaging, and novel liquid biopsy technologies, helping to detect cancers earlier, faster and to give a more accurate diagnosis.
3. Baek S, He Y, Allen BG, Buatti JM, Smith BJ, Tong L, et al. Deep segmentation networks predict survival of non-small cell lung cancer. Sci Rep [Internet]. 2019 Dec 21 [cited 2019 Dec 5];9(1):17286. Available from: http://www.nature.com/articles/s41598-019-53461-2
4. Evison M, Barrett E, Cheng A, Mulla A, Walls G, Johnston D, et al. Predicting the Risk of Disease Recurrence and Death Following Curative-intent Radiotherapy for Non-small Cell Lung Cancer: The Development and Validation of Two Scoring Systems From a Large Multicentre UK Cohort. Clin Oncol. W.B. Saunders; 2021 Mar 1;33(3):145–54.
5. The study also used routinely available clinical data such as age, gender, smoking status, tumour size, shape and type, and intensity of radiotherapy.
6. For example, one of the radiomic features used in this study has been correlated with hypoxia-related carbonic anhydrase (CAIX), a protein that is associated with radiation resistance and poor survival outcomes for NSCLC patients. As patients whose cancer expresses the protein are potentially at a higher risk of recurrence, clinicians could use this technology to tailor their post-treatment follow-up.