Related Experiment Video
Updated: Sep 6, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Automated methods for sella turcica segmentation on cephalometric radiographic data using deep learning (CNN)
Kaushlesh Singh Shakya1,2, Amit Laddi2, Manojkumar Jaiswal3
1Academy of Scientific and Innovative Research (AcSIR), Ghaziabad, 201002, India.
Oral Radiology
|June 23, 2022
Summary
This study introduces a novel deep learning technique for automatic sella turcica segmentation in cephalometric radiographs. VGG19 and ResNet34 CNNs demonstrated superior performance for segmenting this complex anatomical structure.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Accurate segmentation of the sella turcica (ST) is crucial for orthodontic and craniofacial analyses.
- Traditional methods for ST segmentation can be time-consuming and subjective.
- Deep learning offers a promising avenue for automating complex image segmentation tasks.
Purpose of the Study:
- To develop and evaluate novel convolutional neural network (CNN) architectures for the automatic segmentation of the sella turcica (ST) in cephalometric radiographs.
- To compare the performance of different deep learning models for distinguishing ST in complex radiographic images.
Main Methods:
- A dataset of 525 lateral cephalometric images was utilized.
- Ground truth annotations were created by dental specialists using an online platform.
- The study compared fine-tuned CNN architectures: VGG19, ResNet34, InceptionV3, and ResNext50.
Main Results:
- VGG19 achieved a mean IoU of 0.7651 and a dice coefficient of 0.7794.
- ResNet34 achieved a mean IoU of 0.7241 and a dice coefficient of 0.7487.
- InceptionV3 and ResNext50 showed significantly lower performance with mean IoU and dice coefficients around 0.45.
Conclusions:
- VGG19 and ResNet34 architectures significantly outperformed InceptionV3 and ResNext50 for ST segmentation.
- These findings establish a reference model for future investigations into ST morphology and related anomalies.
