RECORD
Automatic ICD-10 Code Association: A Challenging Task on French Clinical Texts
Computation and Language · Submitted 6 Apr 2023
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim Fair experiments on a Clinical dataset in the French language show that our approach increases the $F_1$-score metric by more than 55\% compared to state-of-the-art results.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Automatically associating ICD codes with electronic health data is a well-known NLP task in medical research. NLP has evolved significantly in recent years with the emergence of pre-trained language models based on Transformers architecture, mainly in the English language. This paper adapts these models to automatically associate the ICD codes. Several neural network architectures have been experimented with to address the challenges of dealing with a large set of both input tokens and labels to be guessed. In this paper, we propose a model that combines the latest advances in NLP and multi-label classification for ICD-10 code association. Fair experiments on a Clinical dataset in the French language show that our approach increases the $F_1$-score metric by more than 55\% compared to state-of-the-art results.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Automatically associating ICD codes with electronic health data is a well-known NLP task in medical research. | background 1.00 | 0.04 | 0.02 | asserted 1.00 |
| s1 | NLP has evolved significantly in recent years with the emergence of pre-trained language models based on Transformers architecture, mainly in the English language. | background 0.99 | 0.04 | 0.02 | asserted 1.00 |
| s2 | This paper adapts these models to automatically associate the ICD codes. | contribution 0.81 | 0.78 | 0.04 | asserted 0.95 |
| s3 | Several neural network architectures have been experimented with to address the challenges of dealing with a large set of both input tokens and labels to be guessed. | method 0.83 | 0.66 | 0.03 | asserted 0.98 |
| s4 | In this paper, we propose a model that combines the latest advances in NLP and multi-label classification for ICD-10 code association. | contribution 0.95 | 0.81 | 0.10 | asserted 0.97 |
| s5 | Fair experiments on a Clinical dataset in the French language show that our approach increases the $F_1$-score metric by more than 55\% compared to state-of-the-art results. | result 1.00 | 0.95 | 0.89 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Automatically associating ICD codes with electronic health data is a well-known NLP task in medical research. NLP has evolved significantly in recent years with the emergence of pre-trained language models based on Transformers architecture, mainly in the English language. This paper adapts these models to automatically associate the ICD codes. Several neural network architectures have been experimented with to address the challenges of dealing with a large set of both input tokens and labels to be guessed. In this paper, we propose a model that combines the latest advances in NLP and multi-label classification for ICD-10 code association. Fair experiments on a Clinical dataset in the French language show that our approach increases the $F_1$-score metric by more than 55\% compared to state-of-the-art results.
Human review
How the literature uses it
What this paper does with the arXiv papers it cites
What it cites · 7 sentences without a confident label
Uses the method or tool of 4 papers
-
To overcome the problem of a large set of labels since ICD-10-HNFC contains more than 6,000 codes, we used the Label-Aware Attention (LAAT) system as in [ref] .
PLM-ICD: Automatic ICD Coding with Pretrained Language Models · Classification of a Large Number of Labels · 1.00
-
To overcome this problem, the authors used the Label-Aware Attention (LAAT) mechanism proposed in [ref] which integrates labels in the encoding of documents.
A Label Attention Model for ICD Coding from Clinical Text · ICD Code Association · 0.99
-
As previously exposed, as we deal with the French language our choice was to fine-tune pre-trained transformer-based French models i.e. CamemBERT [ref] and FlauBERT [ref] for the implementation of the model architecture.
CamemBERT: a Tasty French Language Model · Model Architecture · 0.92
Compares with 1 paper
-
We compare the results of recent works (PLM-ICD [ref] , CNN [ref] ) on the association of ICD-10 codes with
PLM-ICD: Automatic ICD Coding with Pretrained Language Models · Experiments and Analysis · 0.99
Cites as background 9 papers
-
The most precise known model is Longformer of [ref] , which is composed of a local attention (attention between a window of neighbour (consecutive) tokens) and a global attention that reduces the computational complexity of the model.
Longformer: The Long-Document Transformer · Sparse-Attention Transformers · 0.84
-
Several strategies have been proposed in the literature [ref] .
Revisiting Transformer-based Models for Long Document Classification · Long sequence processing · 0.80
-
On the other hand, [ref] and [ref] have used the attention mecanism with RNNs and CNNs to implement more accurate models.
Towards Automated ICD Coding Using Deep Learning · ICD Code Association · 0.83
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.