Identifying an Optimal Machine Learning Generated Image Marker to Predict Survival of Gastric Cancer Patients
Published in Proceedings of SPIE, Medical Imaging 2022: Computer-Aided Diagnosis, 2022
This paper introduces a non-iterative cluster routing technique tailored for capsule networks, aiming to overcome the limitations of traditional iterative routing methods like dynamic routing or EM routing.
The proposed method:
- Preserves the part–whole relationship fundamental to capsule networks.
- Eliminates the need for repeated iterations, reducing computational complexity.
- Demonstrates better generalization to novel viewpoints and poses in vision tasks.
- Achieves competitive or superior performance on datasets such as MNIST and CIFAR-10, while using fewer parameters.
The authors provide both theoretical analysis and practical implementation strategies, validating the approach through empirical experiments and architectural comparisons
Recommended citation: Pham, H., Jones, M., Gai, T., Islam, W., Danala, G., & Zheng, B. (2022). "Identifying an Optimal Machine Learning Generated Image Marker to Predict Survival of Gastric Cancer Patients." In Proceedings of SPIE, Vol. 12033, Medical Imaging 2022: Computer-Aided Diagnosis. https://doi.org/10.1117/12.2611788
Download Paper