COMPUTER AIDED FEATURE EXTRACTION IN 3D ORO-MAXILLO-FACIAL IMAGES

Aim: Recently, maxillofacial imaging has drastically evolved thanks to the development of dedicated imaging techniques. The complex anatomy of the maxillofacial region requires the use of different image modalities, demanding the development of dedicated image analysis procedures. This doctoral thesis aims at explore the possible applications of image processing techniques for maxillofacial applications. Hard tissue imaging: In this chapter, we propose different studies that focused on image segmentation, registration and artifact reduction applied to the automatic extraction of hard tissue structures in CBCT data. The procedure involves an adaptive, cluster-based segmentation of bone tissues followed by an intensity-based registration of an annotated reference volume onto a patient CBCT head volume. Automatic segmentation shows a high accuracy level with no significant difference between automatically and manually determined threshold values. The overall median localization error was equal to 1.99 mm with an interquartile range (IQR) of 1.22-2.89 mm. The second study aims to objectively compare the influence of different image parameters on metal artifact generation. After a fully automatic segmentation and image registration, the effect on metal object segmentation and background image noise was evaluated. Results showed that metal object segmentation is highly influenced by the device and material factor, while background noise was more affected by the devices and the FOV parameters compared to the used material. Then, in the third study, in order to improve the automatic extraction of craniofacial features and cephalometric landmarks, we designed a metal artifacts reduction (MAR) algorithm. The new MAR step is fully integrated with our landmark detection algorithm and works on both projection and image domain and allows the automatic detection of the corrupted portion of the image, thus preserving image details. The algorithm was tested on 17 CBCT volumes with a total number of 245 analysed VOIs and reduction of SD values in not metallic voxels was used as image metric. In our dataset, the proposed MAR algorithm always decreased the voxel intensity SD in the examined VOIs, thus showing a significant metal artifact reduction in a fully automatic way. Finally, in order to improve the segmentation quality in the mandibular condyle region, which is usually affected by large amount of noise, a dedicated segmentation approach was developed. Also this algorithm is integrated with the hard tissue segmentation approach and it is based on patient adaptive thresholding and contrast enhancement techniques. The algorithm was tested in vitro on a series of CT and CBCT scans of a dried human mandible. To reproduce soft-tissue attenuation, a cupper filter was used. The proposed automatic segmentation algorithm allows to improve the quality of the trabecular bone segmentation, significantly reducing the overestimation of the segmented bone. Soft tissue imaging: In this section, the application of stereophotogrammetric systems and laser scanners for the development of computer aided approaches for facial morphology evaluation was evaluated. In the first study, we present a new quantitative method to assess symmetry in different facial thirds, objectively defined on trigeminal distribution branches territories. Seventy subjects (40 healthy controls and 30 patients affected by monolateral facial palsy) were acquired with a stereophotogrammetric system and the level of asymmetry was evaluated, RMSD was used as asymmetry metric. Results show a high average reproducibility of area selection and significant differences in RMSD values between controls and patients for all the thirds. No significant differences were found on different thirds among controls, while significant differences were found for upper, middle and lower thirds of patients. The proposed method provides an accurate, reproducible and local facial symmetry analysis. The second study aims to develop an imaging technique that allows to integrate the information about patient dentition together with the stereophotogrammetric reconstruction of the face, providing an un-invasive way to assess the morphology of facial soft tissues in relation to teeth. The proposed algorithm is based on several surface registration steps, initialized by a landmark registration step. To validate the proposed method, CBCT images were analysed and a series of dentofacial distances were calculated. The high values of percentage of corresponding point and a median distance of 0.59 mm prove the accuracy of the registration progress. Statistical analysis shows no significant differences between distanced calculated on CBCT image or on face and dental surfaces, except for one distance. Upper airway imaging: In this last section, a new image analysis method was developed to assess whether three-dimensional morphometric parameters could be useful in nasal septal deviation (NSD) diagnosis and, secondarily, whether CBCT could be considered an adequate imaging technique for the proposed task. Forty-six CBCT scan were segmented using ITK-Snap in order to obtain the 3D model of patient upper airway and compute four morphological parameters septal deviation angle (SDA), percentage of volume difference between right and left side of the nasal airways, nasal airway total volume and a new synthetic septal deviation index (SDI). Principal component analysis (PCA) was used to unveil relationships between each variable and the global nasal airway variability. Among the analysed parameters, SDI seemed to be the most suitable for the quantitative ssessment of NSD, and CBCT allowed accurate assessment of airway morphology. Conclusion: In conclusion, we proved that the application of image processing techniques may help in the development of new diagnostic tools. This PhD thesis has helped creating a good basis for future studies on the application of imaging techniques in oral and maxillofacial applications.