Towards Semi-Structured Automatic ICD Coding via Tree-based Contrastive Learning

Authors: Chang Lu, Chandan Reddy, Ping Wang, Yue Ning

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experimental results demonstrate that our proposed training strategies effectively enhance the performance of existing ICD coding methods.
Researcher Affiliation Academia Department of Computer Science, Stevens Institute of Technology Department of Computer Science, Virginia Tech
Pseudocode Yes Algorithm 1: Section Title Extraction
Open Source Code Yes The code of the proposed DF-IAPF method and training strategies can be found at: https://github.com/Lu Chang-CS/semi-structured-icd-coding.
Open Datasets Yes The MIMIC-III [11] dataset is a popular publicly available EHR dataset
Dataset Splits Yes The detailed training/dev/test dataset statistics for each task are listed in Table 1.
Hardware Specification Yes All programs are executed using a machine with Python 3.9.3, CUDA 11.7, an Intel i9-11900K CPU, 64GB memory, and an NVIDIA RTX 3090 GPU.
Software Dependencies Yes All programs are executed using a machine with Python 3.9.3, CUDA 11.7, an Intel i9-11900K CPU, 64GB memory, and an NVIDIA RTX 3090 GPU.
Experiment Setup Yes For contrastive pre-training, the batch size is 16, the learning rate is 5 10 4, the optimizer is Adam W, and the epoch number is 20. For the masked section training, we set γ to 0.2 for MIMIC-full prediction and 0.3 for MIMIC-50/MIMIC-rare-50 prediction.