BACKGROUND
Lung cancer is the leading cause of cancer death globally, with non-small cell lung cancer (NSCLC) being the most common subtype. Accurate diagnosis of genetic mutations is crucial for selecting appropriate treatments. Artificial Intelligence (AI), particularly deep learning models, shows timely performance in predicting mutations in higher accuracy.
OBJECTIVE
To explore and summarize data on predicting genetic mutations in NSCLC using deep learning models.
METHODS
Four databases were searched without publication time limits, including all studies that used deep learning models to predict genetic mutations in NSCLC patients. Exclusion criteria were studies that used machine learning without deep learning, conference papers, non-English publications, and studies with purposes other than genetic mutation prediction. Screening and data extraction were facilitated by the Rayyan AI tool.
RESULTS
Out of 265 studies identified, 24 met our inclusion criteria. Most studies originated from China and the USA, focusing predominantly on EGFR mutations, followed by KRAS, ALK, and PD-L1. Convolutional Neural Networks (CNNs), especially the ResNet-50 architecture, were the most utilized models. CT scans were the primary data source, with performance metrics like AUC and accuracy ranging from 0.595 to 0.993 and 0.678 to 0.95, respectively.
CONCLUSIONS
EGFR mutation was the mostly identified mutation, having CNN as the predominant model used. The performance of deep learning models in predicting NSCLC mutations shows variability and requires further enhancement. Further studies need to be conducted to accurately evaluate the deep learning performance in predicting the genetic mutation type before these models can be used in clinical practice.
CLINICALTRIAL
Registered in PROSPERO under the following ID: CRD42023489629