Sukmoon Chang

Automatic detection and segmentation of ground glass opacity nodules

Ground Glass Opacity (GGO) is defined as hazy increased attenuation within a lung that is not associated with obscured underlying vessels. Since pure (nonsolid) or mixed (partially solid) GGO at the thin-section CT are more likely to be malignant than those with solid opacity, early detection and treatment of GGO can improve a prognosis of lung cancer. However, due to indistinct boundaries and inter- or intra-observer variation, consistent manual detection and segmentation of GGO have proved to be problematic. In this paper, we propose a novel method for automatic detection and segmentation of GGO from chest CT images. For GGO detection, we develop a classifier by boosting k-NN whose distance measure is the Euclidean distance between the nonparametric density estimates of two examples. The detected GGO region is then automatically segmented by analyzing the texture likelihood map of the region. We applied our method to clinical chest CT volumes containing 10 GGO nodules. The proposed method detected all of the 10 nodules with only one false positive nodule. We also present the statistical validation of the proposed classifier for GGO detection as well as very promising results for automatic GGO segmentation. The proposed method provides a new powerful tool for automatic detection as well as accurate and reproducible segmentation of GGO.

VASCULAR STRUCTURE SEGMENTATION AND BIFURCATION DETECTION

Delineation and reconstruction of vascular structures in medical images are critical for the diagnosis of various vascular diseases and related surgical procedures. In this paper, we present a novel method for vascular structure segmentation with fully automatic detection of bifurcation points. First, we perform a preselection of tubular objects and trace the vessels based on the eigenanalysis of the Hessian matrix. This provides us the estimated direction of vessels as well as the cross-sectional planes orthogonal to the vessels. Then, we apply AdaBoost learning method with specially designed filters on cross-sectional planes to automatically detect the bifurcation points of the vessels. Our method has over 97% success rate for detecting bifurcation points. We present very promising results of our method applied to the reconstruction of pulmonary vessels from clinical chest CT. Our method allows for fully automatic detection of bifurcation points as well as segmentation of vessels

Pulmonary Micronodule Detection from 3D Chest CT

Computed Tomography (CT) is one of the most sensitive medical imaging modalities for detecting pulmonary nodules. Its high contrast resolution allows the detection of small nodules and thus lung cancer at a very early stage. In this paper, we propose a method for automating nodule detection from high-resolution chest CT images. Our method focuses on the detection of discrete types of granulomatous nodules less than 5 mm in size using a series of 3D filters. Pulmonary nodules can be anywhere inside the lung, e.g., on lung walls, near vessels, or they may even be penetrated by vessels. For this reason, we first develop a new cylinder filter to suppress vessels and noise. Although nodules usually have higher intensity values than surrounding regions, many malignant nodules are of low contrast. In order not to ignore low contrast nodules, we develop a spherical filter to further enhance nodule intensity values, which is a novel 3D extension of Variable N-Quoit filter. As with most automatic nodule detection methods, our method generates false positive nodules. To address this, we also develop a filter for false positive elimination. Finally, we present promising results of applying our method to various clinical chest CT datasets with over 90% detection rate.

An Automatic Method for Ground Glass Opacity Nodule Detection and Segmentation from CT Studies

Ground glass opacity (GGO) is defined as hazy increased attenuation within a lung that is not associated with obscured underlying vessels. Since pure (non-solid) or mixed (partially solid) GGO at the thin-section CT are more likely to be malignant than those with solid opacity, early detection and treatment of GGO can improve a prognosis of lung cancer. However, due to indistinct boundaries and inter-or intra-observer variation, consistent manual detection and segmentation of GGO have proved to be problematic. In this paper, we propose a novel method for automatic detection and segmentation of GGO from chest CT images. For GGO detection, we develop a classifier by boosting k-nearest neighbor (k-NN), whose distance measure is the Euclidean distance between the nonparametric density estimates of two regions. The detected GGO region is then automatically segmented by analyzing the 3D texture likelihood map of the region. We applied our method to clinical chest CT volumes containing 10 GGO nodules. The proposed method detected all of the 10 nodules with only one false positive nodule. We also present the statistical validation of the proposed classifier for automatic GGO detection as well as very promising results for automatic GGO segmentation. The proposed method provides a new powerful tool for automatic detection as well as accurate and reproducible segmentation of GGO

Sound source localization for robot auditory systems

Sound source localization (SSL) is a major function of robot auditory systems for intelligent home robots. The steered response power-phase transform (SRP-PHAT) is a widely used method for robust SSL. However, it is too slow to run in real time, since SRP-PHAT searches a large number of candidate sound source locations. This paper proposes a search space clustering method designed to speed up the SRP-PHAT based sound source localization algorithm for intelligent home robots equipped with small scale microphone arrays. The proposed method reduces the number of candidate sound source locations by 30.6% and achieves 46.7% error reduction compared to conventional methods.

A voice trigger system using keyword and speaker recognition for mobile devices

Voice activity detection plays an important role for an efficient voice interface between human and mobile devices, since it can be used as a trigger to activate an automatic speech recognition module of a mobile device. If the input speech signal can be recognized as a predefined magic word coming from a legitimate user, it can be utilized as a trigger. In this paper, we propose a voice trigger system using a keyword-dependent speaker recognition technique. The voice trigger must be able to perform keyword recognition, as well as speaker recognition, without using computationally demanding speech recognizers to properly trigger a mobile device with low computational power consumption. We propose a template based method and a hidden Markov model (HMM) based method for the voice trigger to solve this problem. The experiments using a Korean word corpus show that the template based method performed 4.1 times faster than the HMM based method. However, the HMM based method reduced the recognition error by 27.8% relatively compared to the template based method. The proposed methods are complementary and can be used selectively depending on the device of interest.

Scan-Conversion Algorithm for Ridge Point Detection on Tubular Objects

Anatomical structures contain various types of curvilinear or tube-like structures such as blood vessels and bronchial trees. In medical imaging, the extraction and representation of such structures are of clinical importance. Complex curvilinear structures can be best represented by their center lines (or skeletons) along their elongated direction. In this paper, a gradient-based method for ridge point extraction on tubular objects is presented. Using the gradients of distance maps or intensity profiles usually generates skeleton surfaces for 3D objects, which is not desirable for representing tubular objects. To extract only the points on the centerline, we first employ the gradient vector flow (GVF) technique and then apply eigenanalysis of the Hessian matrix to remove false positive points. We present various results of the method using CLSM (Confocal Laser Scanning Microscopy) images of blood fibrins and CT images of a skull and lungs. Our method is efficient and allows for completely automatic extraction of points along the centerline of a tubular object in its elongated direction.

Vessel boundary extraction using ridge scan-conversion deformable model

Delineation and reconstruction of curvilinear structures in medical images are critical for the diagnosis of various vascular diseases and related surgical procedures. In this paper, we present a novel method for vascular structure segmentation and reconstruction, including the automatic detection of bifurcation points. First, we perform a preselection of tubular structures. Second, we trace the vessels based on the eigenanalysis of the Hessian matrix. This provides us the estimated direction of vessels as well as the cross-sectional planes orthogonal to the vessels. A ScanConversion method is then applied to cross-sectional planes to automatically detect the bifurcation points of the vessels. This method has a 96.59% success rate for detecting bifurcation correctly. Finally, vessels are delineated and reconstructed using deformable models. Our method is efficient and allows for completely automatic delineation and reconstruction of vessels as well as automatic detection of bifurcation points

An efficient audio fingerprint search algorithm for music retrieval

The conventional audio fingerprinting system by Haitsma uses a lookup table to identify the candidate songs in the database, which contains the sub-fingerprints of songs, and searches the candidates to find a song whose bit error rate is the lowest. However, this approach has a drawback that the number of database accesses increases dramatically, especially when the database contains a large number of songs or when a matching sub-fingerprint is not found in the lookup table due to a heavily degraded input signal. In this paper, a novel search method is proposed to overcome these difficulties. The proposed method partitions each song found from the lookup table into blocks, assigns a weight to each block, and uses the weight as a search priority to speed up the search process while reducing the number of database accesses. Various results from our experiment show the significant improvement in search speed while maintaining the search accuracy comparable to the conventional method.

A novel learning based segmentation method for rodent brain structures using MRI

This paper reports a novel method for fully automated segmentation of rodent brain volume by extending the robust active shape models to incorporate an automatic prior shape selection process. This automatic prior shape selection process using support vector machines provides an automatic shape initialization method for further segmentation of rodent brain structures such as Cerebellum, Neocortex, Corpus Callosum, External Capsule, Caudate Putamen, Hippocampus and Ventricles with the robust active shape model framework in magnetic resonance images (MRI). The mean successful rate of this classification method shows 92.2% accuracy compared to the expert-defined ground truth. We also demonstrate the very promising segmentation results of the robust active shape model framework in rodent brain volume.