FaceAutoPipeline

This repository provides a pipeline for fully automated processing of head CT/MRI scans and detection of nose tips from face point-cloud and face mesh reconstructed automatically from DICOM images.

Script Overview

• DICOM_reader.py
  Reads a DICOM dataset, extracts image metadata, and reconstructs the image volume as a 3D matrix.

• DeepMVLM.py
  Extracts facial landmarks from a 3D head/face mesh.
  ⚠️ Note: To use this script, download the github repo of the model here: https://github.com/RasmusRPaulsen/Deep-MVLM and make sure to place it inside the directory containing these scripts. ⚠️ Note: Inside Deep-MVLM/configs, edit the DTU3D-depth-MRI.json file in this way:

  "pre-align": {
    	"align_center_of_mass" : true,
    	"rot_x": -90,
    	"rot_y": 0,
    	"rot_z": 0,
    	"scale": 1,
    	"write_pre_aligned": true
    }
  

• Head_extraction.py
  Automatically segments the patient's head using VTK libraries.

• pointcloud_aligner.py
  Contains a class for point cloud registration using ICP with perturbation to improve convergence.

• pointcloud_nose_tip.py
  Detects the nose tip from a point cloud using a geometric heuristic.

• Main.py
  Run this script to perform head segmentation, nose tip detection, and registration between the face point cloud and the head isosurface.
  📁 The directory path to set in the code should be the one containing the DICOM dataset.

• requirements.txt
  Contains the libraries required

License & Citations

If you use this project or the scripts contained within, please cite:

@misc{Albanesi2025FaceAutoPipeline,
  author       = {Alessandro Albanesi},
  title        = {FaceAutoPipeline},
  howpublished = {\url{https://github.com/aless98/FaceAutoPipeline}},
  year         = {2025}
}

A. Albanesi, FaceAutoPipeline. 2025. [Online]. Available: https://github.com/aless98/FaceAutoPipeline