Research suggests an artificial intelligence (AI) tool called DeepGEM may provide an advancement in genomic testing that offers an accurate, cost-effective, and timely method for gene mutation prediction from histopathology slides. The research was presented at the International Association for the Study of Lung Cancer (IASLC) 2024 World Conference on Lung Cancer by Wenhua Liang, MD, of the China State Key Laboratory of Respiratory Disease and National Clinical Research Center for Respiratory Disease, the First Affiliated Hospital of Guangzhou Medical University, China.
Background
Accurate detection of driver gene mutations is essential for effective treatment planning and prognosis prediction in lung cancer. Traditional genomic testing, which relies on high-quality tissue samples, is often time-consuming and resource-intensive, limiting accessibility, particularly in low-resource settings. Addressing this gap, Prof. Liang and colleagues used DeepGEM, which uses routinely acquired histology slides to predict gene mutations, significantly enhancing accessibility and efficiency in mutation screening.
Prof. Liang and his colleagues analyzed data sets from 16 centers and 3,658 patients. This data set, which includes paired pathologic images and gene mutation data, was complemented by publicly available data sets from The Cancer Genome Atlas.
DeepGEM was initially trained and evaluated on an internal data set of 1,717 patients and subsequently tested on an external data set from 15 additional centers with 1,719 patients and a public data set of 535 patients. To assess generalizability, the model was also tested on a lymph node metastases data set consisting of 331 biopsies.
“DeepGEM also provides interpretable results, generating spatial gene mutation maps at the single-cell level, which have been validated against immunohistochemistry results, underscoring the model’s accuracy and reliability,” Prof. Liang said.
Results
DeepGEM demonstrated performance with a median area under the curve (AUC) of 0.938 for excisional biopsies and 0.891 for aspiration biopsies in the internal data set. On the multicenter external data set, the model achieved median area under the curve (AUC) of 0.859 for excisional biopsies and 0.826 for aspiration biopsies. The model was further validated with an AUC of 0.874 on the TCGA data set, showcasing its effectiveness across diverse racial backgrounds. Importantly, DeepGEM’s ability to predict mutations from primary biopsies extended to lymph node metastases, indicating its potential for prognostic prediction of targeted therapies.
According to Prof. Liang, DeepGEM also provides interpretable results, generating spatial gene mutation maps at the single-cell level, which have been validated against immunohistochemistry results, underscoring the model’s accuracy and reliability.
Summary
“Compared to previous studies, DeepGEM achieved robust and superior predictive performance across various genes validating on the largest multicentre datasets to date. The rapid prediction capabilities of DeepGEM allow for quicker decision-making in treatment, enabling patients with severe symptoms to receive targeted therapies sooner. Furthermore, it presents opportunities for multigene mutation detection and precision treatment in economically underdeveloped areas where genomic testing is unaffordable. This innovative approach has the potential to transform the clinical management of lung cancer patients, making advanced genomic insights more accessible and actionable,” Prof. Liang reported.
