Skip to Main content Skip to Navigation
Master Thesis

Classification automatique de comptes rendus anatomo-pathologiques en texte libre : application au sein d’un registre de cancers

Abstract : Pathology reports are ones of the main information sources concerning a cancer register. They are unstructured textual data, their content can be sum up by adding a describing code of one or several medical concepts. When these data are not coded, they need a manual processing in order to extract the wanted information. Our aim is to automate this not coded pathology reports’ processing among french cancer registers through automatic textual categorization models by learning machine. Based on 84 745 pathology reports’ extracts from the Gironde general cancer register’ data, we created three categorization models and evaluated various choices according to: their data, representation, and algorithms of categorization used. The evaluation measures were: the recall, the precision and the F-measure. For the binary classification model, allowing to distinguish the pathology reports relating to a cancer, we get a 0.964 F-measure. For the cancer topography transcription, the best model gets a 0.703 micro F-measure and a 0.633 macro F-measure. These results are relevant and suggest the possibility of using those models to automate in free texts written in French, the pathology reports’ processing. Nevertheless, improvements are necessary to let their production start within a cancer register.
Document type :
Master Thesis
Complete list of metadatas

Cited literature [63 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-02407000
Contributor : Isped Ub <>
Submitted on : Thursday, December 12, 2019 - 12:12:11 PM
Last modification on : Friday, October 23, 2020 - 4:38:39 PM

File

M2SITIS_2019_NAFFRECHOUX.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : dumas-02407000, version 1

Collections

Citation

Benjamin Naffrechoux. Classification automatique de comptes rendus anatomo-pathologiques en texte libre : application au sein d’un registre de cancers. Santé publique et épidémiologie. 2019. ⟨dumas-02407000⟩

Share

Metrics

Record views

72

Files downloads

110