Fusing Data from CT Deep Learning, CT Radiomics and Peripheral Blood Immune profiles to Diagnose Lung Cancer in a Cohort of Patients Experiencing Symptoms
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Background
Lung cancer is the leading cause of cancer-related deaths. Diagnosis at late stages is common due to the largely non-specific nature of presenting symptoms contributing to high mortality. There is a lack of specific, minimally invasive low-cost tests to screen patients ahead of the diagnostic biopsy.
Methods
344 patients experiencing symptoms from the lung clinic of Lister hospital suspected of lung cancer were recruited. Predictive covariates were successfully generated on 170 patients from Computed Tomography (CT) scans using CT Texture Analysis (CTTA) and Deep Learning Autoencoders (DLA) as well as from peripheral blood data for immunity using high depth flow-cytometry and for exosome protein components. Predictive signatures were formed by combining covariates using Bayesian regression on a randomly chosen 128-patient training set and validated on a 42-patient held-out set. Final signatures were generated by fusing the data sources at different levels.
Findings
Immune, CTTA and DLA single modality signatures had overall AUCs of 0.69, 0.70 and 0.73 respectively. The final combined signature had a ROC AUC of 0.81. The overall sensitivity and specificity were 0.72 and 0.77 respectively.
Interpretation
Combining immune monitoring with CT scan data is an effective approach to improving sensitivity and specificity of Lung cancer screening even in patients experiencing symptoms.
Funding
CRUK [C1519/A27375], Wellcome Trust/EPSRC Centre for Medical Engineering [WT203148/Z/16/Z], NIHR Clinical Research Facility at Guy’s and St Thomas’ NHS Foundation Trust, NIHR Biomedical Research Centre.
Research in Context
Evidence before this study
Lung cancer is the leading cause of cancer related deaths and previous studies have shown that early diagnosis is often difficult and one third of patients return to their GP three or more times before a referral to a specialist according to a study in the British Journal of Cancer in 2013. Screening is a possible solution and low dose CT scans for patients with high-risk has been implemented, but this excludes non-smokers and the young and can be inaccurate according to studies in 2022 and 2023. In addition, the workload of screening on radiologists is resulting in delays in examinations as recently reported by the Royal College of Radiologists in 2024. Some studies have suggested the benefit of a blood test, to augment a CT scan, based on DNA fragments or circulating tumour cells, for example (2021-2023).
A PubMed publications database search (search term: “blood“[Title] AND “CT“[Title] AND “Lung“[Title] AND “cancer“[Title]) revealed 17 results on 8 th May 2025, of which 2 were directly relevant involving combining a PET/CT scan with inflammatory blood measurements, rejecting those where blood referred to blood volume or perfusion imaging or ctDNA.
Added value of this study
We propose a blood test based on the immune response to the presence of cancer (in contrast to other studies which are based on detecting nucleic acids) that is combined with the automated analysis of CT scans (reducing radiologist workload) by machine learning, which may provide a route to screening patients with symptoms of lung diseases. The advantage of a blood test based on the immune response is that immune cell-based detection of cancers can occur when the tumours are relatively small and at an earlier time point than the shedding of nucleic acids. We recruited 344 patients between October 2020 and November 2021 into the LungExoDETECT study, with a blood test for immune profiling and routine CT scan. After following their progress to determine their cancer diagnosis, we were able to make a mathematical model that predicts diagnosis from 12 measurements based on the blood and CT scan, and which was validated in a patient subset.
Implications of all the available evidence
Our study has important implications for the field: (1) We reinforce previous our previous evidence that the response of the immune system to cancer can add additional information for detecting cancer to that provided by the automated analysis of CT scans. (2) Advanced Bayesian machine learning can produce simple mathematical models that predict the presence of cancer and are highly interpretable in terms of why a particular prediction has been made by the model by pinpointing the particular component of the immune system that contributes to the prediction.
