The face as a window to the brain: Development of a pipeline for facial segmentation from magnetic resonance scans

Puyuelo Citoler, Alba

The face as a window to the brain: Development of a pipeline for facial segmentation from magnetic resonance scans

Citation

Puyuelo Citoler, A. The face as a window to the brain: Development of a pipeline for facial segmentation from magnetic resonance scans. 2022. handle: http://hdl.handle.net/10230/54141

This citation was generated automatically.

Abstract

The biological evidence that links facial and brain development gives rise to the long-standing hypothesis on the relation between facial geometry and some neuropsychiatric disorders. However, the existence of studies directly comparing face and brain data is still scarce. Thus, this study was part of a research project that aimed to detect whether morphological brain abnormalities linked to schizophrenia may also be related to abnormalities in the face geometry. A novel database of Magnetic Resonance scans of the head of schizophrenia and control patients was used to explore the existence of this link. The role of this study in the project was to develop the pipeline to extract meaningful facial characteristics. Twenty head scans were processed, ten of which were considered outliers due to their high levels of noise. Consequently, the project focused on eliminating the noise present in the volumes without losing geometry details. The pipeline is formed by an initial pre-processing block to eliminate part of the mentioned noise, followed by facial region segmentation through several pixel intensity-based methods. Finally, as noise remained in some cases, two public 3D Morphable Head Models were employed to deform a template head mesh extracted from these models to the target heads obtained in the segmentation, followed by a mesh refinement of the deformed head. In this way, all the subjects were brought into correspondence, having the same number of vertices and triangles for further statistical analysis. This study presents the different methods applied in the pre-processing and segmentation steps, the parameters tested in the two statistical methods, and the methods tested in the mesh refinement, all with their corresponding results. Finally, it outlines the ones that gave results with enough quality for the extraction of facial features. Thereby, this approach could be employed as an initial step in the research project's goal to detect the possible facial dysmorphogenesis linked to brain abnormalities and schizophrenia.