Automatic segmentation of ameloblastoma on ct images using deep learning with limited data

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Ameloblastoma is a common benign jaw bone tumor requiring precise segmentation for diagnosis and treatment.
• Manual segmentation methods for ameloblastoma are inefficient and have drawbacks.
• AI-based automatic segmentation offers a crucial improvement for clinical workflows.

Purpose of the Study:

• To develop and evaluate an AI-based automatic segmentation approach for ameloblastoma using deep learning.
• To enhance the accuracy and efficiency of ameloblastoma diagnosis and treatment planning.
• To assess the generalization performance of the proposed model on external datasets.

Main Methods:

• Collected CT images from 79 ameloblastoma patients.
• Utilized a Mask R-CNN deep learning neural network architecture.
• Employed image preprocessing, enhancement, and cross-validation techniques for model training and testing.
• Validated the model using an external dataset of 200 CT images.

Main Results:

• The AI model successfully performed automatic ameloblastoma segmentation.
• Achieved a DICE index of 0.874.
• Demonstrated strong performance with an Average Precision (AP) of 0.741 (IoU 0.5-0.95), 0.914 (IoU 0.5), and 0.826 (IoU 0.75).
• External validation confirmed robust generalization capabilities.

Conclusions:

• A deep learning-based neural network effectively segments ameloblastoma automatically.
• The proposed method significantly improves efficiency, accuracy, and speed in clinical settings.
• This AI approach is a promising tool for enhancing ameloblastoma diagnosis and treatment.