Ayse Betul Oktay

Localization of the Lumbar discs using machine learning and exact probabilistic inference

We propose a novel fully automatic approach to localize the lumbar intervertebral discs in MR images with PHOG based SVM and a probabilistic graphical model. At the local level, our method assigns a score to each pixel in target image that indicates whether it is a disc center or not. At the global level, we define a chain-like graphical model that represents the lumbar intervertebral discs and we use an exact inference algorithm to localize the discs. Our main contributions are the employment of the SVM with the PHOG based descriptor which is robust against variations of the discs and a graphical model that reflects the linear nature of the vertebral column. Our inference algorithm runs in polynomial time and produces globally optimal results. The developed system is validated on a real spine MRI dataset and the final localization results are favorable compared to the results reported in the literature.

Simultaneous Localization of Lumbar Vertebrae and Intervertebral Discs With SVM-Based MRF

This paper presents a method for localizing and labeling the lumbar vertebrae and intervertebral discs in mid-sagittal MR image slices. The approach is based on a Markov-chain-like graphical model of the ordered discs and vertebrae in the lumbar spine. The graphical model is formulated by combining local image features and semiglobal geometrical information. The local image features are extracted from the image by employing pyramidal histogram of oriented gradients (PHOG) and a novel descriptor that we call image projection descriptor (IPD). These features are trained with support vector machines (SVM) and each pixel in the target image is locally assigned a score. These local scores are combined with the semiglobal geometrical information like the distance ratio and angle between the neighboring structures under the Markov random field (MRF) framework. An exact localization of discs and vertebrae is inferred from the MRF by finding a maximum a posteriori solution efficiently using dynamic programming. As a result of the novel features introduced, our system can scale-invariantly localize discs and vertebra at the same time even in the existence of missing structures. The proposed system is tested and validated on a clinical lumbar spine MR image dataset containing 80 subjects of which 64 have disc- and vertebra-related diseases and abnormalities. The experiments show that our system is successful even in abnormal cases and our results are comparable to the state of the art.

Computer aided diagnosis of degenerative intervertebral disc diseases from lumbar MR images.

This paper presents a novel method for the automated diagnosis of the degenerative intervertebral disc disease in midsagittal MR images. The approach is based on combining distinct disc features under a machine learning framework. The discs in the lumbar MR images are first localized and segmented. Then, intensity, shape, context, and texture features of the discs are extracted with various techniques. A Support Vector Machine classifier is applied to classify the discs as normal or degenerated. The method is tested and validated on a clinical lumbar spine dataset containing 102 subjects and the results are comparable to the state of the art.

A Novel Level Set Based Echocardiographic Contour Extraction Method with Prior Knowledge.

This paper presents a novel level set method with shape priors. The method keeps the level set deformations and the integration of the prior information as separate processes and hence it can be used with any level set formulation without complicating the level set functional. The method does not need any explicit training phase and by the addition of an appropriate deformable contour matching method, it can be used for any specific application. The system is tested and verified by the task of extraction of the inner and outer heart walls (endocardium and epicardium) from the echocardiographic images of the left ventricle.

Echocardiographic contour extraction with local and global priors through boosting and level sets

Extracting endocardium and epicardium from echocardiographic images is a challenging task because of large amounts of noise, signal drop-out, unrelated structures, and unseen wall parts. This paper introduces a new technique that automatically extracts cardiac borders by incorporating local and global priors through boosting and level set methods. The shape-based global prior is incorporated into the system by regularly re-initializing the level set surface under the influence of the expert detected contours. The local priors with image and temporal information are learned through boosting. The proposed system has many advantages. First, boosting encodes the knowledge about the image information and the temporal cardiac wall motion effectively by using spatiotemporal filters. Second, the local priors can use any features from the images including different filters and intensity profiles. Furthermore, other hard constraints like local shape, texture, distance, etc. can be added to local features effectively. The system is validated on echocardiograms and the results are found to be promising.

Prostate detection from abdominal ultrasound images: A part based approach

Prostate cancer is one of the most frequent cancers among men. Abdominal ultrasound imaging is a very practical alternative to more precise but inconvenient transrectal ultrasound imaging for the diagnosis and treatment of prostate cancer. However, detection of the prostate region alone is very difficult for the abdominal ultrasound images. This paper presents a new prostate detection method that models the abdominal images as the classes of neighboring anatomical regions of the prostate. The proposed method has two levels: Pixel level detection assigns class scores to each pixel in the image. Model level detection uses these scores to determine the final positions of the anatomical regions in the image. The proposed approach is very effective for the specific problems of the abdominal ultrasound scans. Extensive experiments performed on real patient data with and without pathologies produce very promising results.

Prior information based segmentation: A 3D level set surface matching approach

This paper presents a level set based segmentation method with shape priors. The shape priors guide the level set deformations so that the contour extraction process is affected not only from the local image properties, but also from the expert knowledge in the form of manual contours. The method does not need an explicit training phase and it does not complicate the level set functional because level set deformations and incorporation of prior information are done separately. The system uses manual expert contours to produce new level set surfaces which are warped into the surface from the level set process. The prior information is incorporated into the level sets by re-initializing these warped surfaces as new level set surfaces. The resulting system is validated by running experiments on synthetic data and real MR and ultrasound cardiac images.

Automatic Tooth Identification in Dental Panoramic Images with Atlas-based Models

After catastrophes and mass disasters, accurate and efficient identification of decedents requires an automatic system which depends upon strong biometrics. In this paper, we present an automatic tooth detection and labeling system based on panoramic dental radiographs. Although our ultimate objective is to identify decedents by comparing the postmortem and antemortem dental radiographs, this paper only involves the tooth detection and the tooth labeling stages. In the system, the tooth regions are first determined and the detection module runs for each region individually. By employing the sliding window technique, the Haar features are extracted from each window and the SVM classifies the windows as tooth or not. The labeling module labels the candidate tooth positions determined by the SVM with an atlas-based model and the final tooth positions are inferred. The novelty of our system is combining the atlas-based model with the SVM under the same framework. We tested our system on 35 panoramic images and the results are promising.

Segmentation of Fe 3 O 4 nano particles in TEM images

Automatic segmentation of nanoparticles and determination of their shapes and sizes from transmission electron microscopy images are crucial for material analysis. Manual segmentation of nanoparticles produces subjective results and it is time-consuming. In this study, a new method is proposed for the automatic segmentation of the nanoparticles. First, background and foreground detection is employed with machine learning. Then, the nanoparticles are coarsely detected with connected component analysis and they are determined with Hough Transform. The method is tested on ten different images. The nanoparticles segmented with our method are similar to the nanoparticles segmented manually by experts and ImageJ software and the results are promising.

Desiccation detection from lumbar MR images

Desiccation is the drying out of the fluids in the lumbar intervertebral discs and it may cause many health problems. In clinical practice, MR imaging is used for diagnosis because in T2-weighted MR images the desiccated discs are darker than non-desiccated discs. In this study, we present a method for automatically detecting desiccated lumbar intervertebral discs from MR images. First, the lumbar discs are automatically localized and labeled. Then, raw intensity features are used and texture features are extracted with local binary patterns technique from the lumbar discs. The features are trained and tested by random forests. The method is tested and validated on a dataset containing 80 MR images. The classification accuracy of the method is %88.54 and results are promising.