Deep learning approach for disease detection in lumbosacral spine radiographs using ConvNet

Lumbosacral exam enables several computer-aided diagnoses (CAD) applications since it provides detailed spine radiographs, specifically of the lumbar, sacral, and coccygeal regions. We developed a deep learning-based CAD system methodology to detect lumbosacral anomalies in this context. For this pu...

Full description

Saved in:
Bibliographic Details
Published inComputer methods in biomechanics and biomedical engineering. Vol. 11; no. 6; pp. 2560 - 2575
Main Authors de Abreu Vieira, Pablo, Vogado, Luis, Lopes, Lucas, Ricardo, Ricardo, Santos Neto, Pedro, Mathew, Mano Joseph, Magalhães, Deborah, Silva, Romuere
Format Journal Article
LanguageEnglish
Published Taylor & Francis 02.11.2023
Subjects
Computer Science
Deep learning
Lumbosacral
Medical Imaging
Spine
Transfer-learning
X-Ray
Deep learning
X-Ray
Lumbosacral
Spine
Transferlearning
medical imaging
Online AccessGet full text

Cover

More Information
Summary:Lumbosacral exam enables several computer-aided diagnoses (CAD) applications since it provides detailed spine radiographs, specifically of the lumbar, sacral, and coccygeal regions. We developed a deep learning-based CAD system methodology to detect lumbosacral anomalies in this context. For this purpose, we used a heterogeneous dataset that contains frontal and lateral lumbosacral exams. Our classification experiments achieved promising results with pre-trained CNNs: accuracy of 0.82/0.86, kappa of 0.65/0.66, and F1-Score of 0.82/0.83 for frontal and lateral images, respectively. In addition, we developed an ensemble classification that combines frontal and lateral images of the same exam and uses a confidence threshold selection. In this ensemble, we managed to issue alerts of 22.31% with a False Discovery Rate of 2.79%. Our proposal allows specialists to assist in decision-making.
ISSN:2168-1163
2168-1171
DOI:10.1080/21681163.2023.2245922