MICA: Towards Explainable Skin Lesion Diagnosis via Multi-Level Image-Concept Alignment
Authors: Yequan Bie, Luyang Luo, Hao Chen
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental results on three skin image datasets demonstrate that our method, while preserving model interpretability, attains high performance and label efficiency for concept detection and disease diagnosis.In Table 1, we report the classification comparison results of our method under three metrics (AUROC, Accuracy and F1 Score) on the considered datasets. |
| Researcher Affiliation | Academia | 1Department of Computer Science and Engineering, Hong Kong University of Science and Technology 2Department of Chemical and Biological Engineering, Hong Kong University of Science and Technology 3HKUST Shenzhen-Hong Kong Collaborative Innovation Research Institute |
| Pseudocode | No | The paper describes the overall framework in Section 3 and illustrates it with Figure 2, but it does not contain any formal pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/Tommy-Bie/MICA. |
| Open Datasets | Yes | Datasets: Derm7pt (Kawahara et al. 2018) is a dermoscopic image dataset... PH2 (Mendonc a et al. 2013) contains a total of 200 dermoscopic images... Skin Con (Daneshjou et al. 2022) is a skin disease dataset... |
| Dataset Splits | Yes | The dataset is split into training set, validation set and test set according to the proportion of 70%, 15% and 15%, respectively. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions using ResNet-50, BERT encoder, and Adam optimizer but does not specify the version numbers for the software libraries or programming languages used for implementation, such as PyTorch, TensorFlow, or Python versions. |
| Experiment Setup | Yes | We adopt Adam (Kingma and Ba 2014) optimizer with learning rate of 5e-5 in the first stage and 1e-4 in the second stage. For the hyperparameter selection, we use grid search and set τ1 = 0.25, τ2 = 0.2, τ3 = 0.1. We set β = 1 for Derm7pt and Skin Con dataset, and set β = 0.5 for PH2 dataset. |