README.md

Back to Projects List

Automatic Landmark Identification in IOS and Cranio Facial CBCT

Key Investigators

• Maxime Gillot (UoM)
• Baptiste Baquero (UoM)
• Jonas Bianchi (UoM, UoP)
• Marcela Gurge (UoM)
• Najla Al Turkestani (UoM)
• Marilia Yatabe (UoM)
• Lucia Cevidanes (UoM)
• Juan Prieto (UoNC)

Project Description

For CBCT : We propose a novel approach that reformulates anatomical landmark detection as a classification problem through a virtual agent placed inside a 3D Cone-Beam Computed Tomography (CBCT) scan. This agent is trained to navigate in a multi-scale volumetric space to reach the estimated landmark position. The agent movements decision relies on a combination of Densely Connected Convolutional Networks (DCCN) and fully connected layers.

For IOS :

Objective

1. Combine the old ALI CBCT of the first project week and the new ALI IOS in a new module
2. Add new landmarks in the available list
3. Deploy the tool in a Slicer module

Approach and Plan

1. Use the module we worked on during project week 36
2. Get new data to train on with more landmarks

Progress and Next Steps

1. We have models for landmark identification in CBCT and IOS
2. We have the begenning of an UI on slicer
3. The next steps are :
4. Link the UI with both ALI IOS and ALI CBCT algorithms
5. Train new models for more landmarks
6. Deploy the tool as a Slicer module in the sclicer CMF extention

---

1. I havn't been able to make any progress on this module during this week as I've been working on AMASSS CBCT (Maxime Gillot, Baptiste Baquero,Lucia Cevidanes, Juan Prieto) and Slicer Batch Annonymize (Hina Shah, Juan Carolos Prieto)
2. I learned how to use CLI modules that will make the development/deployment of ALI faster in the near future.
3. Nes agents have been trained to reach a total of 120 landmarks that can be automatically identified

Illustrations

Background and References