The Use of Computer Vision for Localization of Vertebrae on Midsagittal Computed Tomography Slices

Automation of routine operations associated with medical image analysis is an important problem because it reduces the workload of radiologists. The selection of computed tomography (CT) images of vertebral levels for body composition assessment is generally performed by hand, which requires additio...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Programming and computer software Ročník 51; číslo 6; s. 385 - 394
Hlavní autori: Teplyakova, A. R., Shershnev, R. V.
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Moscow Pleiades Publishing 01.12.2025
Springer Nature B.V
Predmet:
Accuracy
Algorithms
Artificial Intelligence
Body composition
Computed tomography
Computer Science
Computer vision
Data augmentation
Estimates
Image analysis
Image segmentation
Localization
Lung cancer
Medical imaging
Modules
Operating Systems
Preprocessing
Skips
Software
Software Engineering
Software Engineering/Programming and Operating Systems
Tomography
Vertebrae
Visualization
Workloads
computed tomography
localization
medical images
deep learning
preprocessing
U-Net
computer vision
vertebra
segmentation
augmentation
ISSN:0361-7688, 1608-3261
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Automation of routine operations associated with medical image analysis is an important problem because it reduces the workload of radiologists. The selection of computed tomography (CT) images of vertebral levels for body composition assessment is generally performed by hand, which requires additional time. The purpose of this study is to develop an approach to solving the problem of vertebrae localization on midsagittal CT slices for automated selection of axial slices to assess body composition. The developed approach is based on a multiclass segmentation model with the U-Net family architecture and computer vision methods for image preprocessing and segmentation mask postprocessing. To estimate the effect of input data types and model architectures on segmentation accuracy, we consider 20 different configurations of the approach. It is found that the proposed method for input data preprocessing, based on generation of three-channel images, improves the accuracy of multiclass segmentation for four out of five architectures (Dense U-Net achieves the maximum Dice similarity coefficient of 0.8858). It is also found that the proposed training data augmentation method, which skips axial slices when forming sagittal slices, improves the multiclass segmentation accuracy for models with the ResU-Net and Dense U-Net architectures. Based on the proposed approach, a software module is implemented for automatic detection of the cervical, thoracic, and lumbar vertebrae on midsagittal CT slices, their visualization, and determination of axial slice indices that correspond to vertebral body centers. The developed module is integrated into a program for visualization and analysis of DICOM files. The developed module can be used as an auxiliary tool in solving diagnostic problems.
Bibliografia:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0361-7688
1608-3261
DOI:10.1134/S0361768825700252