Three dimensional study to quantify the relationship between facial hard and soft tissue movement as a result of orthognathic surgery

Introduction Prediction of soft tissue changes following orthognathic surgery has been frequently attempted in the past decades. It has gradually progressed from the classic “cut and paste” of photographs to the computer assisted 2D surgical prediction planning; and finally, comprehensive 3D surgical planning was introduced to help surgeons and patients to decide on the magnitude and direction of surgical movements as well as the type of surgery to be considered for the correction of facial dysmorphology. A wealth of experience was gained and numerous published literature is available which has augmented the knowledge of facial soft tissue behaviour and helped to improve the ability to closely simulate facial changes following orthognathic surgery. This was particularly noticed following the introduction of the three dimensional imaging into the medical research and clinical applications. Several approaches have been considered to mathematically predict soft tissue changes in three dimensions, following orthognathic surgery. The most common are the Finite element model and Mass tensor Model. These were developed into software packages which are currently used in clinical practice. In general, these methods produce an acceptable level of prediction accuracy of soft tissue changes following orthognathic surgery. Studies, however, have shown a limited prediction accuracy at specific regions of the face, in particular the areas around the lips. Aims The aim of this project is to conduct a comprehensive assessment of hard and soft tissue changes following orthognathic surgery and introduce a new method for prediction of facial soft tissue changes.   Methodology The study was carried out on the pre- and post-operative CBCT images of 100 patients who received their orthognathic surgery treatment at Glasgow dental hospital and school, Glasgow, UK. Three groups of patients were included in the analysis; patients who underwent Le Fort I maxillary advancement surgery; bilateral sagittal split mandibular advancement surgery or bimaxillary advancement surgery. A generic facial mesh was used to standardise the information obtained from individual patient’s facial image and Principal component analysis (PCA) was applied to interpolate the correlations between the skeletal surgical displacement and the resultant soft tissue changes. The identified relationship between hard tissue and soft tissue was then applied on a new set of preoperative 3D facial images and the predicted results were compared to the actual surgical changes measured from their post-operative 3D facial images. A set of validation studies was conducted. To include: • Comparison between voxel based registration and surface registration to analyse changes following orthognathic surgery. The results showed there was no statistically significant difference between the two methods. Voxel based registration, however, showed more reliability as it preserved the link between the soft tissue and skeletal structures of the face during the image registration process. Accordingly, voxel based registration was the method of choice for superimposition of the pre- and post-operative images. The result of this study was published in a refereed journal. • Direct DICOM slice landmarking; a novel technique to quantify the direction and magnitude of skeletal surgical movements. This method represents a new approach to quantify maxillary and mandibular surgical displacement in three dimensions. The technique includes measuring the distance of corresponding landmarks digitized directly on DICOM image slices in relation to three dimensional reference planes. The accuracy of the measurements was assessed against a set of “gold standard” measurements extracted from simulated model surgery. The results confirmed the accuracy of the method within 0.34mm. Therefore, the method was applied in this study. The results of this validation were published in a peer refereed journal. • The use of a generic mesh to assess soft tissue changes using stereophotogrammetry. The generic facial mesh played a major role in the soft tissue dense correspondence analysis. The conformed generic mesh represented the geometrical information of the individual’s facial mesh on which it was conformed (elastically deformed). Therefore, the accuracy of generic mesh conformation is essential to guarantee an accurate replica of the individual facial characteristics. The results showed an acceptable overall mean error of the conformation of generic mesh 1 mm. The results of this study were accepted for publication in peer refereed scientific journal. Skeletal tissue analysis was performed using the validated “Direct DICOM slices landmarking method” while soft tissue analysis was performed using Dense correspondence analysis. The analysis of soft tissue was novel and produced a comprehensive description of facial changes in response to orthognathic surgery. The results were accepted for publication in a refereed scientific Journal. The main soft tissue changes associated with Le Fort I were advancement at the midface region combined with widening of the paranasal, upper lip and nostrils. Minor changes were noticed at the tip of the nose and oral commissures. The main soft tissue changes associated with mandibular advancement surgery were advancement and downward displacement of the chin and lower lip regions, limited widening of the lower lip and slight reversion of the lower lip vermilion combined with minimal backward displacement of the upper lip were recorded. Minimal changes were observed on the oral commissures. The main soft tissue changes associated with bimaxillary advancement surgery were generalized advancement of the middle and lower thirds of the face combined with widening of the paranasal, upper lip and nostrils regions. In Le Fort I cases, the correlation between the changes of the facial soft tissue and the skeletal surgical movements was assessed using PCA. A statistical method known as ’Leave one out cross validation’ was applied on the 30 cases which had Le Fort I osteotomy surgical procedure to effectively utilize the data for the prediction algorithm. The prediction accuracy of soft tissue changes showed a mean error ranging between (0.0006mm±0.582) at the nose region to (-0.0316mm±2.1996) at the various facial regions.

The role of context in image annotation and recommendation

With the rise of smart phones, lifelogging devices (e.g. Google Glass) and popularity of image sharing websites (e.g. Flickr), users are capturing and sharing every aspect of their life online producing a wealth of visual content. Of these uploaded images, the majority are poorly annotated or exist in complete semantic isolation making the process of building retrieval systems difficult as one must firstly understand the meaning of an image in order to retrieve it. To alleviate this problem, many image sharing websites offer manual annotation tools which allow the user to “tag” their photos, however, these techniques are laborious and as a result have been poorly adopted; Sigurbjörnsson and van Zwol (2008) showed that 64% of images uploaded to Flickr are annotated with < 4 tags. Due to this, an entire body of research has focused on the automatic annotation of images (Hanbury, 2008; Smeulders et al., 2000; Zhang et al., 2012a) where one attempts to bridge the semantic gap between an image’s appearance and meaning e.g. the objects present. Despite two decades of research the semantic gap still largely exists and as a result automatic annotation models often offer unsatisfactory performance for industrial implementation. Further, these techniques can only annotate what they see, thus ignoring the “bigger picture” surrounding an image (e.g. its location, the event, the people present etc). Much work has therefore focused on building photo tag recommendation (PTR) methods which aid the user in the annotation process by suggesting tags related to those already present. These works have mainly focused on computing relationships between tags based on historical images e.g. that NY and timessquare co-exist in many images and are therefore highly correlated. However, tags are inherently noisy, sparse and ill-defined often resulting in poor PTR accuracy e.g. does NY refer to New York or New Year? This thesis proposes the exploitation of an image’s context which, unlike textual evidences, is always present, in order to alleviate this ambiguity in the tag recommendation process. Specifically we exploit the “what, who, where, when and how” of the image capture process in order to complement textual evidences in various photo tag recommendation and retrieval scenarios. In part II, we combine text, content-based (e.g. # of faces present) and contextual (e.g. day-of-the-week taken) signals for tag recommendation purposes, achieving up to a 75% improvement to precision@5 in comparison to a text-only TF-IDF baseline. We then consider external knowledge sources (i.e. Wikipedia & Twitter) as an alternative to (slower moving) Flickr in order to build recommendation models on, showing that similar accuracy could be achieved on these faster moving, yet entirely textual, datasets. In part II, we also highlight the merits of diversifying tag recommendation lists before discussing at length various problems with existing automatic image annotation and photo tag recommendation evaluation collections. In part III, we propose three new image retrieval scenarios, namely “visual event summarisation”, “image popularity prediction” and “lifelog summarisation”. In the first scenario, we attempt to produce a rank of relevant and diverse images for various news events by (i) removing irrelevant images such memes and visual duplicates (ii) before semantically clustering images based on the tweets in which they were originally posted. Using this approach, we were able to achieve over 50% precision for images in the top 5 ranks. In the second retrieval scenario, we show that by combining contextual and content-based features from images, we are able to predict if it will become “popular” (or not) with 74% accuracy, using an SVM classifier. Finally, in chapter 9 we employ blur detection and perceptual-hash clustering in order to remove noisy images from lifelogs, before combining visual and geo-temporal signals in order to capture a user’s “key moments” within their day. We believe that the results of this thesis show an important step towards building effective image retrieval models when there lacks sufficient textual content (i.e. a cold start).