Automated Risser Grade Assessment of Pelvic Bones Using Deep Learning

(1) Background: This study aimed to develop a deep learning model using a convolutional neural network (CNN) to automate Risser grade assessment from pelvic radiographs. (2) Methods: We used 1619 pelvic radiographs from patients aged 12–18 years with scoliosis to train two CNN models—one for the rig...

Full description

Saved in:
Bibliographic Details
Main Authors: Jeoung Kun Kim, Donghwi Park, Min Cheol Chang
Format: Article
Language:English
Published: MDPI AG 2025-05-01
Series:Bioengineering
Subjects:
Online Access:https://www.mdpi.com/2306-5354/12/6/589
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849433579193368576
author Jeoung Kun Kim
Donghwi Park
Min Cheol Chang
author_facet Jeoung Kun Kim
Donghwi Park
Min Cheol Chang
author_sort Jeoung Kun Kim
collection DOAJ
description (1) Background: This study aimed to develop a deep learning model using a convolutional neural network (CNN) to automate Risser grade assessment from pelvic radiographs. (2) Methods: We used 1619 pelvic radiographs from patients aged 12–18 years with scoliosis to train two CNN models—one for the right pelvis and one for the left. A multimodal approach incorporated 224 × 224-pixel regions of interest from radiographs, alongside patient age and gender. The models were optimized with Adam, weight decay, rectified linear unit (ReLU) activation, dropout, and batch normalization, while synthetic data augmentation addressed class imbalance. Performance was evaluated through accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (ROC AUC). (3) Results: The right pelvis model achieved 83.64% accuracy; the left pelvis model reached 80.56%. Both models performed well for Risser Grades 0, 2, and 4, with the right pelvis model achieving a microaverage F1-score of 0.836 and ROC AUC of 0.895. The left pelvis model achieved a microaverage F1-score of 0.806 and ROC AUC of 0.872. Challenges arose from class imbalance in less frequent grades. (4) Conclusions: CNN models effectively automated Risser grade assessment, reducing clinician workload and variability.
format Article
id doaj-art-e659b80559d94f46b8a47898b9a096f4
institution Kabale University
issn 2306-5354
language English
publishDate 2025-05-01
publisher MDPI AG
record_format Article
series Bioengineering
spelling doaj-art-e659b80559d94f46b8a47898b9a096f42025-08-20T03:26:57ZengMDPI AGBioengineering2306-53542025-05-0112658910.3390/bioengineering12060589Automated Risser Grade Assessment of Pelvic Bones Using Deep LearningJeoung Kun Kim0Donghwi Park1Min Cheol Chang2Department of Business Administration, School of Business, Yeungnam University, Gyeongsan-si 38541, Republic of KoreaSeoul Spine Rehabilitation Clinic, Ulsan-si 44607, Republic of KoreaDepartment of Rehabilitation Medicine, College of Medicine, Yeungnam University, Daegu 42415, Republic of Korea(1) Background: This study aimed to develop a deep learning model using a convolutional neural network (CNN) to automate Risser grade assessment from pelvic radiographs. (2) Methods: We used 1619 pelvic radiographs from patients aged 12–18 years with scoliosis to train two CNN models—one for the right pelvis and one for the left. A multimodal approach incorporated 224 × 224-pixel regions of interest from radiographs, alongside patient age and gender. The models were optimized with Adam, weight decay, rectified linear unit (ReLU) activation, dropout, and batch normalization, while synthetic data augmentation addressed class imbalance. Performance was evaluated through accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (ROC AUC). (3) Results: The right pelvis model achieved 83.64% accuracy; the left pelvis model reached 80.56%. Both models performed well for Risser Grades 0, 2, and 4, with the right pelvis model achieving a microaverage F1-score of 0.836 and ROC AUC of 0.895. The left pelvis model achieved a microaverage F1-score of 0.806 and ROC AUC of 0.872. Challenges arose from class imbalance in less frequent grades. (4) Conclusions: CNN models effectively automated Risser grade assessment, reducing clinician workload and variability.https://www.mdpi.com/2306-5354/12/6/589Risser gradepelvic bonebone ageradiographdeep learningartificial intelligence
spellingShingle Jeoung Kun Kim
Donghwi Park
Min Cheol Chang
Automated Risser Grade Assessment of Pelvic Bones Using Deep Learning
Bioengineering
Risser grade
pelvic bone
bone age
radiograph
deep learning
artificial intelligence
title Automated Risser Grade Assessment of Pelvic Bones Using Deep Learning
title_full Automated Risser Grade Assessment of Pelvic Bones Using Deep Learning
title_fullStr Automated Risser Grade Assessment of Pelvic Bones Using Deep Learning
title_full_unstemmed Automated Risser Grade Assessment of Pelvic Bones Using Deep Learning
title_short Automated Risser Grade Assessment of Pelvic Bones Using Deep Learning
title_sort automated risser grade assessment of pelvic bones using deep learning
topic Risser grade
pelvic bone
bone age
radiograph
deep learning
artificial intelligence
url https://www.mdpi.com/2306-5354/12/6/589
work_keys_str_mv AT jeoungkunkim automatedrissergradeassessmentofpelvicbonesusingdeeplearning
AT donghwipark automatedrissergradeassessmentofpelvicbonesusingdeeplearning
AT mincheolchang automatedrissergradeassessmentofpelvicbonesusingdeeplearning