Skip to Main content Skip to Navigation
Master Thesis

Intérêt de la réduction de dimension pour évaluation de la fibrose hépatique en échographie : approche radiomique basée sur les contours du foie

Abstract : Purpose: Hepatic fibrosis assessment is a challenge for chronic liver disease medical care. The goal of this study was to investigate the performance of a dimensionality reduction model to assess significant hepatic fibrosis, based on radiomics extracted from ultrasonographic (US) images of liver surface. Methods: ln this retrospective study, 114 patients with a liver biopsy for liver disease or disorder, or a clinically diagnosed cirrhosis, had an US examination of the liver surface. After liver edge segmentation and radiomics extraction, images were analyzed using dimensionality reduction algorithm UMAP (Uniform Manifold Approximation and Projection) to distinguish non-significant (group G0 = F0-1) from significant fibrosis (G1 = F2-4), with probability density maps. Using area under receiver operating characteristic curve (AUROC) estimation, results were compared to the expert radiologist subjective reading, to transient elastography (Fibroscan®) and to a home software for automated surface irregularities assessment. Results: 109 patients (G0: n0=56; G1: n1=53) were analyzed using UMAP. Using 3 dimensions reduction, global precision of unsupervised learning classification was 67% (G0: 69%; G1: 64%). AUROC was 0.76, better than the expert interpretation (0.67, p=0.02), but not different from Fibroscan® (0.8, p=0.483) or from the software (0.8, p=0.17). Conclusion: Dimensionality reduction classification allows performances that are not different from the state-of-the-art classification techniques. UMAP density maps establish an atlas to which non-radiologic data can be added, in order to develop a predictive tool of hepatic fibrosis evolution.
Document type :
Master Thesis
Complete list of metadatas

Cited literature [62 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-02954552
Contributor : Jean-Hugues Morneau <>
Submitted on : Thursday, October 1, 2020 - 10:46:16 AM
Last modification on : Wednesday, October 14, 2020 - 4:13:21 AM

File

2019GREA5146_teyssier_yann(1)(...
Files produced by the author(s)

Identifiers

  • HAL Id : dumas-02954552, version 1

Citation

Yann Teyssier. Intérêt de la réduction de dimension pour évaluation de la fibrose hépatique en échographie : approche radiomique basée sur les contours du foie. Médecine humaine et pathologie. 2019. ⟨dumas-02954552⟩

Share

Metrics

Record views

27

Files downloads

9