Automatic Segmentation of Morphological Structures, Metastasis Detection, and 3D Model Reconstruction from Medical Imaging Utilising Artificial Intelligence Based on Deep Neural Network Methodologies

Objectives
Deep learning has significantly advanced medical imaging. The application of AI-based methods to tasks such as metastasis detection, segmentation of interstitial cells of Cajal (ICC), and the generation of anatomically accurate 3D models highlights the potential of these technologies in improving medical workflows. This study integrates findings from multiple projects to evaluate the utility and effectiveness of deep learning in clinical practice and medical education.
This study aims to evaluate the application of deep neural networks in:
1.Automated segmentation of complex anatomical structures.
2.Detection and localization of metastatic lesions in CT scans.
3.Automated counting and analysis of ICCs in histological images.
4.Reconstruction of 3D anatomical models for educational purposes.

Materials and Methods
CT scans of spinal metastases (38 cases) and vertebrae (115 cases) were processed using U-Net based architectures. Mandibular CBCT scans (188 cases) were analyzed using ResNet-101 for osteoporosis detection. The ICC cell analysis relied on a YOLO-based architecture, which accurately segmented and quantified cell distributions. Radiological data were post-processed with 3D Slicer to create 28 3D models.

Results
Segmentation and Detection
The U-Net architecture achieved high accuracy in vertebra segmentation, with DSC values between 0.87 and 0.96. Metastasis detection was more challenging, with DSC values of 0.71 for lytic lesions and 0.61 for sclerotic lesions.
Cell Counting
The AI-based cell counting model achieved high accuracy in identifying ICC distributions in histological samples. Automated cell counts showed strong concordance with manual counts, significantly reducing time required for analysis.
3D Model Reconstruction
A total of 28 anatomically accurate 3D models were reconstructed and printed. These models can be used in educational setting to improve student understanding of complex anatomical structures.

Conclusions
Deep learning models effectively automate critical tasks in medical imaging, including segmentation, detection, and cell counting. Integrating artificial intelligence-driven technologies into medical education connects theoretical knowledge with practical skills, promoting innovation and improving results.