Tumor-CapsNet

Reproduce results from "Brain Tumor Type Classification via Capsule Networks" (Afshar et al.) for a final project in a Deep Learning class at Cal Poly San Luis Obispo.

---

Requirements

• Keras v2.0+ (?)
• NumPy v1.13+
• Code from CapsNet-Keras

Dataset

Brain Tumor MR Data posted by Jun Cheng

Architecture (WIP)

• Original presentation of capsule networks for MNIST done in 2017 by Sabour et al. [3]
• Going to reproduce results done by Afshar et al. [1] and get within at least 0.5% of their results (86.56%).
• May use deconvolutional capsule networks proposed by LaLonde and Bagci [2].

Bibliography

1. Brain Tumor Type Classification via Capsule Networks by Afshar, Mohammadi, Plataniotis
2. Capsules for Object Segmentation by LaLonde, Bagci
3. Dynamic Routing Between Capsules by Sabour, Frosst, Hinton