ParseCaps: An Interpretable Parsing Capsule Network for Medical Image Diagnosis

Authors: Xinyu Geng, Jiaming Wang, Xiaolin Huang, Fanglin Chen, Jun Xu

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on three medical datasets show that Parse Caps not only outperforms other capsule network variants in classification accuracy and robustness, but also provides interpretable explanations, regardless of the availability of concept labels.
Researcher Affiliation Academia 1Harbin Institute of Technology, Shenzhen 2Shanghai Jiaotong University EMAIL, EMAIL, EMAIL, EMAIL
Pseudocode Yes Algorithm 1: SAA Routing
Open Source Code Yes 1The code is released at https://github.com/ornamentt/Parsecaps. The supplementary material is available at https://arxiv.org/pdf/2411.01564.
Open Datasets Yes We evaluated Parse Caps with Contrast Enhanced Magnetic Resonance Images (CE-MRI) (Cheng 2017), PH2 (Mendonc a et al. 2013) and Derm7pt (D7) (Kawahara et al. 2019) datasets.
Dataset Splits Yes we split all datasets into 80% training, 10% testing, and 10% validation.
Hardware Specification Yes Parse Caps was developed in Py Torch 12.1 and Python 3.9, accelerated by eight GTX-3090 GPUs.
Software Dependencies Yes Parse Caps was developed in Py Torch 12.1 and Python 3.9, accelerated by eight GTX-3090 GPUs.
Experiment Setup Yes We set the learning rate to 2.5e-3, batch size to 64, and weight decay to 5e-4. The model was trained for 300 epochs using the Adam W optimizer and a 5-cycle linear warm-up.