Electrical and Electronic Engineering - Theses

Permanent URI for this collection

Search Results

Now showing 1 - 1 of 1
  • Item
    Thumbnail Image
    Automatic analysis of 4D laryngeal CT scans to assist diagnosing of voice disorders
    Hewavitharanage, Sajini Ruwanthika Gintota ( 2017)
    Vocal folds are the two smooth bands of muscles located in larynx just above the trachea. Humans produce voice by vibrating the vocal folds using the air coming from the lungs. This abduction and adduction of vocal folds are controlled by the muscles connected to thyroid cartilage, cricothyroid cartilages and arytenoid cartilages. When vocal muscles are misused or excessively used, they can be strained or damaged and voice disorders may occur. Furthermore, vocal folds can be damaged and the connecting cartilages and muscles can be affected due to the effect of other illnesses like Parkinson's disease (PD), multiple sclerosis (MS), myasthenia gravis (MG), strokes or tumours. PD is a neuro-degenerative disease which currently has neither cure nor any pathological tests to detect. The disease progresses very slowly over the years and symptoms appear when approximately 70% of the neuron cells have ceased to function. Usual symptoms are tremors and stiffness in the body muscles which results in difficulty moving most of the body parts externally as well as internally. Consequently, vocal folds and laryngeal muscles get affected and PD patients suffer from vocal impairments. Furthermore, previous studies carried out using laryngoendoscopy, laryngostroboscopy and laryngeal electromyography of PD patients found that those patients have an abnormal phase closure and abnormal laryngeal muscle activity. Moreover, in 2014, a study carried out using a group of early PD patients demonstrated increased glottis area and reduced inter-arytenoid distance in subjects. Therefore, laryngeal measurements could be used as a biomarker for early detection of PD. However, segmenting the vocal folds region from volumetric laryngeal computed tomography (CT) images is a tedious task, when it is done manually. Manual segmentation schemes require lot of expert knowledge and time, and often provide poor objective and reproducible results. In this project, we hope to develop a novel automated algorithm to segment the vocal folds region and measure the laryngeal parameters. This thesis consists of two major parts; first it proposes a fully automated segmentation method for segmenting the rima glottidis from 3D laryngeal CT scans and generates the time series for rima glottidis areas, which in future can be used to develop an automated diagnosis tool for voice disorders. The gray-level difference features are learnt through a support vector machine classifier and several post processing algorithms are introduced to refine the final segmentation result. Second, a fully automated method to estimate the vocal plane position in a 3D laryngeal CT volume using computer vision algorithms and techniques is proposed. Vocal plane position is identified using anatomical markers like thyroid cartilage and vertebral bones and these markers are segmented using gray-scale and edge-based features. The experiments are conducted using a private data set from the Movement Disorder Clinic at Monash Medical Centre The detailed implementations of the two methods including feature extraction, kernel selection, post processing and validation are explained in this thesis.