Title: Medical Coding: Automatic ICD-10 code classification & Transcription Errors in Context of Intent Detection and Filling

Abstract:
(1) Medical coding is the process of identifying diagnoses in medical records and assigning the corresponding ICD-10 codes. A list of codes is compiled for each patient and is reported by hospitals to bill the health care provider. To automate the task, it can be framed as a multi-label classification task. The high amount of different codes and exponential frequency distribution calls for a system combination of a 'deep' and 'wide' model.

(2) Spoken Language Understanding systems typically consist of an Automatic Speech Recognition (ASR) and Natural Language Understanding (NLU) component. Transcription errors during ASR are propagated the NLU component and decrease the performance of intent detection and slot filling. A model jointly trained to correct ASR errors helps to make the NLU module more robust to corrupt transcriptions.