Volume 11, Issue 4 (2-2025)
jhbmi 2025, 11(4): 0-0 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Mohammadian M, Khazaee Moghadam M, Saniei E. Automatic Detection and Segmentation of Bone, lung, and soft tissue Based on Computed Tomography Scan Using Deep learning. jhbmi 2025; 11 (4)
URL: http://jhbmi.ir/article-1-908-en.html
URL: http://jhbmi.ir/article-1-908-en.html
Assistant Professor, Department of Nuclear Engineering, CT.C., Islamic Azad University, Tehran, Iran
Abstract: (606 Views)
Introduction: In the field of radiation therapy and dosimetry, identifying and segmenting different regions and structures of the body in medical images is crucial for calculating radiation dose distribution for optimization, protecting sensitive organs, and improving treatment planning. In this regard, the application of automatic segmentation algorithms and the advancement of deep learning models in the field of medical image analysis enhance both accuracy and speed. This study aims to evaluate two convolutional neural networks (CNNs) and introduce an effective and efficient model for high-accuracy segmentation of bone, lung, and soft tissue based on CT images.
Method: This study utilized deep neural networks based on DeepLabV3+, ResNet-18, and MobileNet-v2 pre-trained architectures as the backbone for segmentation. To preprocess the CT scan images and prepare the input data for the neural network algorithms, 3D Slicer was employed to generate mask images for organs, including soft tissues, bones, and lungs. The development process and fine-tuning of the aforementioned deep learning models were carried out in the MATLAB environment, with accuracy and Intersection over Union (IoU) measured to evaluate the performance of the segmentation algorithms.
Results: The results indicated that the accuracy of semantic segmentation of bone for the ResNet-18 and MobileNet-v2 neural networks was 97% and 96%, respectively. For lung and soft tissue segmentation, the accuracy of the aforementioned networks was reported as 96.9% and 96.7% (for lung), and 99.2% and 99% (for soft tissue), respectively.
Furthermore, the IoU criterion for semantic segmentation of bone by the ResNet-18 and MobileNet-v2 networks was measured at 85% and 84%, respectively. For lung and soft tissue segmentation, this criterion was 90.8% and 91.2% (for lung), and 99% and 99% (for soft tissue) for both networks, respectively.
Conclusion: Various evaluation metrics indicate that the MobileNet-v2 neural network demonstrates superior performance and speed compared to the ResNet-18 network in analyzing CT scan images and segmenting the target tissues.
Method: This study utilized deep neural networks based on DeepLabV3+, ResNet-18, and MobileNet-v2 pre-trained architectures as the backbone for segmentation. To preprocess the CT scan images and prepare the input data for the neural network algorithms, 3D Slicer was employed to generate mask images for organs, including soft tissues, bones, and lungs. The development process and fine-tuning of the aforementioned deep learning models were carried out in the MATLAB environment, with accuracy and Intersection over Union (IoU) measured to evaluate the performance of the segmentation algorithms.
Results: The results indicated that the accuracy of semantic segmentation of bone for the ResNet-18 and MobileNet-v2 neural networks was 97% and 96%, respectively. For lung and soft tissue segmentation, the accuracy of the aforementioned networks was reported as 96.9% and 96.7% (for lung), and 99.2% and 99% (for soft tissue), respectively.
Furthermore, the IoU criterion for semantic segmentation of bone by the ResNet-18 and MobileNet-v2 networks was measured at 85% and 84%, respectively. For lung and soft tissue segmentation, this criterion was 90.8% and 91.2% (for lung), and 99% and 99% (for soft tissue) for both networks, respectively.
Conclusion: Various evaluation metrics indicate that the MobileNet-v2 neural network demonstrates superior performance and speed compared to the ResNet-18 network in analyzing CT scan images and segmenting the target tissues.
Keywords: Semantic segmentation, medical images, computed tomography images, convolutional neural networks (CNN)
Type of Study: Original Article |
Subject:
Artificial Intelligence in Healthcare
Received: 2024/12/7 | Accepted: 2025/03/6
Received: 2024/12/7 | Accepted: 2025/03/6
Audio File [MP3 717 KB] (5 Download)
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
