Deep Learning-Based Anatomical Segmentation of the Foot and Ankle: A Multi-View Radiograph Approach

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Orthopedics

Background:

• Accurate segmentation of foot and ankle bones is crucial for diagnosing various musculoskeletal conditions.
• Traditional manual segmentation methods are time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To develop and evaluate deep learning models for automated identification and segmentation of 14 foot and ankle bones.
• To predict segmentation masks for these bones using multi-view radiographs.

Main Methods:

• Retrospective collection of 273 multi-view radiographs (anteroposterior, oblique, lateral) from 99 patients.
• Manual annotation of ground truth masks for 14 bones (AP/OBL views) and 6 bones (LAT view).
• Fine-tuning of two deep learning models, fully convolutional network (FCN) with ResNet-50 and DeepLabv3 with ResNet-50, for semantic segmentation.

Main Results:

• Both models achieved high mean intersection over union (mIoU) scores, indicating excellent segmentation accuracy.
• FCN-ResNet-50 demonstrated slightly higher mIoU values compared to DeepLabv3 across all views.
• Statistically significant differences in performance were observed between the models, particularly in the oblique view and for individual bone segmentation.

Conclusions:

• Deep learning models, specifically FCN-ResNet-50, are effective for automated segmentation of foot and ankle bone structures.
• These models show potential for improving efficiency and consistency in radiological assessments of the foot and ankle.