HIMS-Net: Horizontal-vertical interaction and multiple side-outputs network for cyst segmentation in jaw images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oral and Maxillofacial Surgery

Background:

• Jaw cysts can cause significant symptoms like facial swelling and tooth loss.
• The complexity and diversity of cyst images present challenges for current deep learning algorithms.
• Accurate segmentation of jaw cysts is crucial for effective diagnosis and treatment planning.

Purpose of the Study:

• To develop and evaluate a novel deep learning network for accurate jaw cyst segmentation.
• To address the limitations of existing algorithms in handling complex cyst image data.
• To improve the speed and accuracy of clinical diagnosis for jaw cysts.

Main Methods:

• Introduction of a horizontal-vertical interaction mechanism for enhanced context dependency capture.
• Utilization of a feature-fused unit to acquire multi-scale contextual information.
• Implementation of a multiple side-outputs strategy for generating detailed segmentation maps.

Main Results:

• The proposed HIMS-Net achieved high performance metrics: Matthews correlation coefficient (Mcc) of 93.61%, Dice of 93.66%, and Jaccard of 88.10%.
• Experimental results on a self-established dataset demonstrated the network's effectiveness.
• Performance was comparable to specialist physicians, indicating clinical usability.

Conclusions:

• HIMS-Net demonstrates significant potential for rapid and accurate jaw cyst diagnosis.
• The network's innovative architecture effectively handles the complexities of jaw cyst image segmentation.
• This advancement could lead to improved patient outcomes through earlier and more precise diagnoses.