jamesmullenbach/caml-mimic

Implements convolutional and attention-based neural architectures (CAML and DR-CAML) for automated ICD code assignment from clinical discharge summaries, with built-in support for both MIMIC-II and MIMIC-III datasets. The approach combines CNNs with per-label attention mechanisms to generate interpretable predictions by highlighting which text passages drive each code decision. Includes end-to-end data processing pipelines, pre-trained model weights, and evaluation scripts for reproducing published benchmarks.

302 stars. No commits in the last 6 months.