An automatic staging approach of early endometrial cancer based on CUDA tree genetic programming

Endometrial cancer (EC) is serious gynecological diseases, and early detection can improve patient prognosis. Staging is important for patients with EC. However, the small-scale images and the complex content of MRI images with EC make staging challenging. Genetic programming (GP) has been applied to image staging on small datasets. Most existing GP methods run on the CPU, making them inefficient. To address this, cuda tree genetic programming (CTGP) is proposed for staging MRI images with EC, running on the GPU. The CTGP includes new feature extraction functions for MRI images with EC. Additionally, this study divides the original EC MRI dataset into three datasets with different training samples based on practical applications. A public dataset is also used to demonstrate the generalization performance of the CTGP. The performance of the CTGP with staging accuracy of 88.5% outperforms the 15 baseline methods, including 5 state-of-the-art GP methods. Furthermore, the CTGP was 6.9, 16.5, and 24.3 times faster than the CPU on three datasets, and 11.1 times faster on the public dataset. In addition, the CTGP offers good interpretability, making it facilitate accurate diagnoses in real clinical scenarios.