Photo Credit: Selvanegra

An AI model was able to outperform a radiomics model in detecting epidermal growth factor receptor gene mutation in patients with NSCLC and brain metastasis.

A deep learning model outperformed a traditional handcrafted radiomics model in preoperatively detecting epidermal growth factor receptor (EGFR) gene mutation in patients with non-small cell lung cancer (NSCLC) and brain metastasis, according to a study published in Quantitative Imaging in Medicine and Surgery.

“Tyrosine kinase inhibitors have been suggested as an effective therapy for patients harboring the EGFR mutation, resulting in satisfactory outcomes with increased survival rates,” explained a research team from Shenyang, China, in the study introduction. “Therefore, early identification of the EGFR mutation is critical for appropriate decision-making regarding individualized treatment strategies.”

The study included a training cohort and a validation cohort of patients with pathologically confirmed brain metastasis from primary NSCLC who underwent contrast-enhanced T1-weighted and T2-weighted magnetic resonance imaging. The training cohort included 160 patients from Liaoning Cancer Hospital and Institute in Shenyang, and the validation cohort included 72 patients from Shengjing Hospital of China Medical University, Shenyang.

The researchers developed a multiscale feature fusion network (MSF-Net) that integrates several types of feature information and employs channel and spatial attention modules. Residual network (ResNet) 50 served as the backbone network.

“MSF-Net is divided into 3 stages, and the introduced spatial attention module in each stage can learn the feature maps specific to each scale to adapt to irregular brain MRI,” the authors wrote. “Subsequently, the shallow and deep branch features are integrated to enable the model to learn more abundant and significant information for classification instead of relying on the previous layer of features.”

The study compared the effectiveness of the MSF-Net model with a conventional model using radiomics, an emerging technique that quantitatively analyzes a large number of features from medical images.

Compared with a handcrafted radiomics model, the MSF-Net model had better diagnostic performance in predicting EGFR mutation and subtypes. The microaveraged area under the curve (AUC) was 0.91 with MSF-Net and 0.80 with radiomics, according to the study. The macroaveraged AUC was 0.90 with MSF-Net and 0.81 with radiomics.

“MSF-Net is an end-to-end model that can be used as a noninvasive tool to identify the EGFR mutation status and subtypes,”the researchers wrote, “thus aiding clinicians in developing personalized treatment plans.”