AI-based pipeline for early screening of lung cancer: integrating radiology, clinical, and genomics data

Background The prognosis of lung carcinoma has changed since the discovery of molecular targets and their specific drugs. Somatic Epidermal Growth Factor Receptor ( EGFR ) mutations have been reported in lung carcinoma, and these mutant proteins act as substrates for targeted therapies. However, in a resource-constrained country like India, panel-based next-generation sequencing cannot be made available to the population at large. Additional challenges such as adequacy of tissue in case of lung core biopsies and locating suitable tumour tissues as a result of innate intratumoral heterogeneity indicate the necessity of an AI-based end -to -end pipeline capable of automatically detecting and learning more effective lung nodule features from CT images and predicting the probability of the EGFR-mutant. This will help the oncologists and patients in resource-limited settings to achieve near-optimal care and appropriate therapy. Methods The EGFR gene sequencing and CT imaging data of 2277 patients with lung carcinoma were included from three cohorts in India and a White population cohort collected from TCIA. Another cohort LIDC-IDRI was used to train the AIPS-Nodule (AIPS-N) model for automatic detection and characterisation of lung nodules. We explored the value of combining the results of the AIPS-N with the clinical factors in the AIPS-Mutation (AIPS-M) model for predicting EGFR genotype, and it was evaluated by area under the curve (AUC). Findings AIPS-N achieved an average AP50 of 70.19% in detecting the location of nodules within the lung region of interest during validation and predicted the score of fi ve lung nodule properties. The AIPS-M machine learning (ML) and deep learning (DL) models achieved AUCs ranging from 0.587 to 0.910. Interpretation The AIPS suggests that CT imaging combined with a fully automated lung-nodule analysis AI system can predict EGFR genotype and identify patients with an EGFR mutation in a cost-effective and non-invasive manner. Copyright (c) 2024 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).