Diagnosis of osteoporosis disease from bone X-ray images with stacked sparse autoencoder and SVM classifier

This paper focuses on the problem of osteoporosis disease diagnosis from bone X-ray images. The proposed approach takes advantage of the deep learning robustness to extract high-level features from low-level image (pixel intensities). However, the diagnosis of osteoporosis confronts two major challe...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2017 International Conference on Advanced Technologies for Signal and Image Processing (ATSIP) S. 1 - 5
Hauptverfasser: Nasser, Yassine, El Hassouni, Mohammed, Brahim, Abdelbasset, Toumi, Hechmi, Lespessailles, Eric, Jennane, Rachid
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 01.05.2017
Schlagworte:
bone X-Ray images
Bones
Deep learning
Feature extraction
feature learning
Osteoporosis
Sociology
Stacked Sparse Autoencoder
Support vector machines
SVM
Training
X-ray imaging
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This paper focuses on the problem of osteoporosis disease diagnosis from bone X-ray images. The proposed approach takes advantage of the deep learning robustness to extract high-level features from low-level image (pixel intensities). However, the diagnosis of osteoporosis confronts two major challenges, the difficulty of distinguishing between osteoporosis and healthy subjects just from the visual inspection of bone X-ray images, and the need of a large-scale and small-size datasets for training the deep networks. In order to separate the Osteoporosis population (OP) from Control cases (CC) our proposed method performs a series of three consecutive steps, namely: 1) preprocessing to enhance the contrast of the image, 2) image subdivision with the sliding window operation and feature extraction with Staked Sparse Autoencoder (SSAE), 3) pooling operation followed by classification step using the SVM classifier. Experimental results indicate that a performance gain on classification of the two populations (OP and CC) was achieved.
DOI:10.1109/ATSIP.2017.8075537