Bald eagle-optimized transformer networks with temporal-spatial mid-level features for pancreatic tumor...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Pancreatic tumor classification is challenging due to complexity and variability.
• Traditional methods lack precision in capturing tumor dynamics.
• Advanced techniques are needed for robust diagnosis.

Purpose of the Study:

• To develop an advanced method for pancreatic tumor classification.
• To integrate temporal-spatial features with optimized transformer networks.
• To improve diagnostic accuracy and efficiency in pancreatic cancer detection.

Main Methods:

• Utilized temporal-spatial mid-level features (CTSF) for enhanced classification.
• Employed bald eagle search (BES) to optimize vision transformer (ViT) and swin transformer (ST) models.
• Evaluated GRU, LSTM, ViT, and ST models on multiple pancreatic CT datasets.

Main Results:

• The ViT model showed superior performance in initial evaluations.
• The swin transformer (ST) achieved high accuracy with spatial features.
• The integrated CTSF model, optimized with BES, reached up to 98.89% accuracy.
• Statistical tests confirmed the superiority of the CTSF-BES approach.

Conclusions:

• Temporal-spatial features are critical for accurate pancreatic tumor classification.
• BES-optimized transformer networks significantly improve diagnostic performance.
• The CTSF-BES approach offers a promising, efficient solution for pancreatic cancer diagnosis.