Deep Learning-based U-Mamba Model to Predict Differentiated Gastric Cancer using Radiomics Features from Spleen...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Manual segmentation of spleen computed tomography (CT) images is time-consuming and prone to inter-observer variability.
• Accurate spleen image analysis is crucial for understanding its role in various diseases.

Purpose of the Study:

• To develop an automated spleen segmentation method using deep learning for computed tomography (CT) images.
• To construct a prediction model for gastric cancer (GC) differentiation using radiomics and clinical data.
• To generate a nomogram for clinical guidance in GC management.

Main Methods:

• A deep learning model (U-Mamba) was used for automated spleen CT image segmentation in 262 patients with confirmed GC.
• Radiomic features were extracted, and dimensionality reduction was applied to identify significant features.
• Three predictive models (CL, RA, CR) were developed by combining clinical and radiomic features, with the best-performing model visualized as a nomogram.

Main Results:

• Thirty radiomic features and one clinical feature were identified as valuable after selection.
• The radiomic features (RA) model showed superior discriminative power compared to the clinical (CL) and combined (CR) models.
• A nomogram was developed based on the logistic clinical model for practical application and validation.

Conclusions:

• Automated spleen segmentation via deep learning enables effective extraction of radiomic features.
• These radiomic features are effective in predicting the differentiation grade of gastric cancer (GC).
• The developed nomogram provides valuable clinical decision-making support for GC.