Transfer Learning-Based Computer Vision Modeling Framework for Lumbar Vertebrae Localization in Sagittal Radiographs

View at publisher Summarise with VIRAT
Authors: Roselle P. Cimagala
Cited by: 0
References: 10

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Accurate localization of lumbar vertebrae in sagittal radiographs is a fundamental prerequisite for quantitative spine analysis, yet manual annotation is laborintensive and prone to inter-observer variability. This study presents a transfer learning-based computer vision framework for automated lumbar vertebrae localization in sagittal lumbar spine X-ray images. The dataset comprises radiographs from a proprietary cohort of Honduran patients with spondylolisthesis and the publicly available BUU-LSPINE dataset, restricted to sagittal views. Vertebrae from L3 to S1 were annotated using bounding boxes to enable supervised vertebral localization. The proposed model leverages pretrained feature representations and multi-scale detection to robustly capture vertebral structures under varying anatomical presentations and imaging conditions. Experimental results demonstrate stable convergence and strong localization performance across the evaluated datasets, with qualitative analyses confirming anatomically consistent vertebral detection. As validation was limited to the available data sources and acquisition settings, further evaluation on larger and more diverse cohorts is necessary to establish broader robustness and generalizability. The findings support the feasibility of automated lumbar vertebral localization and provide a foundation for subsequent quantitative spine analysis.

Fields of Study & Subjects

Medical Imaging and Analysis Advanced Neural Network Applications Dental Radiography and Imaging

Publication details

Pages
1–5
Publication date
24 February 2026
Language
EN