Postdoc Research Scientist, Columbia University
Using artificial intelligence techniques to identify critical disease pattern from heart data
My research focuses on pushing the boundary of data science and technology development to bring to surface the key disease patterns in heart images with the aim of helping clinicians and patients.
Heart disease is the leading cause of human death worldwide. Among various heart disease, atrial fibrillation is one of the most common arrhythmias (irregular heart beating) afflicting approximately 6 million people in the United States. Radio frequency ablation is a standard treatment for cardiac arrhythmias, however ~40 % of atrial fibrillation patients require two procedures to be chronically terminated of AF. The delivery of safe and effective ablation therapy requires precise identification of regions to target and regions to avoid. In addition, the variation of heart architecture among patients highlights the needs for accurate substrate delineation.
I have acquired TBs of human cardiac image data and built a human cardiac atlas based on a high resolution imaging technique, optical coherence tomography. I have employed the artificial intelligence technique, particularly deep learning, to develop a machine learning algorithm that can pore over the large database and automate the classification of diseased tissue regions which are related to irregular heart beating. The developed image analysis tool can potentially facilitate the treatment of irregular heart beating.
Abstract: Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction. These methods have dramatically improved the state-of-the-art in speech recognition, visual object recognition, object detection and many other domains such as drug discovery and genomics. Deep learning discovers intricate structure in large data sets by using the backpropagation algorithm to indicate how a machine should change its internal parameters that are used to compute the representation in each layer from the representation in the previous layer. Deep convolutional nets have brought about breakthroughs in processing images, video, speech and audio, whereas recurrent nets have shone light on sequential data such as text and speech.
Pub.: 29 May '15, Pinned: 31 Aug '17
Abstract: The uterine cervix during pregnancy is the vital mechanical barrier which resists compressive and tensile loads generated from a growing fetus. Premature cervical remodeling and softening is hypothesized to result in the shortening of the cervix, which is known to increase a woman׳s risk of preterm birth. To understand the role of cervical material properties in preventing preterm birth, we derive a cervical material model based on previous mechanical, biochemical and histological experiments conducted on nonpregnant and pregnant human hysterectomy cervical tissue samples. In this study we present a three-dimensional fiber composite model that captures the equilibrium material behavior of the tissue in tension and compression. Cervical tissue is modeled as a fibrous composite material, where a single family of preferentially aligned and continuously distributed collagen fibers are embedded in a compressible neo-Hookean ground substance. The total stress in the collagen solid network is calculated by integrating the fiber stresses. The shape of the fiber distribution is described by an ellipsoid where semi-principal axis lengths are fit to optical coherence tomography measurements. The composite material model is fit to averaged mechanical testing data from uni-axial compression and tension experiments, and averaged material parameters are reported for nonpregnant and term pregnant human cervical tissue. The model is then evaluated by investigating the stress and strain state of a uniform thick-walled cylinder under a compressive stress with collagen fibers preferentially aligned in the circumferential direction. This material modeling framework for the equilibrium behavior of human cervical tissue serves as a basis to determine the role of preferentially-aligned cervical collagen fibers in preventing cervical deformation during pregnancy.
Pub.: 31 Mar '15, Pinned: 31 Aug '17
Abstract: Effects of radiofrequency ablation (RFA) treatment of atrial fibrillation can be limited by the ability to characterize the tissue in contact. Parameters obtained by conventional catheters, such as impedance and temperature can be insufficient in providing physiological information pertaining to effective treatment. In this report, we present a near-infrared spectroscopy (NIRS)-integrated catheter capable of extracting tissue optical properties. Validation experiments were first performed in tissue phantoms with known optical properties. We then apply the technique for characterization of myocardial tissues in swine and human hearts, ex vivo. Additionally, we demonstrate the recovery of critical parameters relevant to RFA therapy including contact verification, and lesion transmurality. These findings support the application of NIRS for improved guidance in RFA therapeutic interventions.
Pub.: 24 Jul '15, Pinned: 31 Aug '17
Abstract: The structural integrity of the cervix in pregnancy is necessary for carrying a pregnancy until term, and the organization of human cervical tissue collagen likely plays an important role in the tissue's structural function. Collagen fibers in the cervical extracellular matrix exhibit preferential directionality, and this collagen network ultrastructure is hypothesized to reorient and remodel during cervical softening and dilation at time of parturition. Within the cervix, the upper half is substantially loaded during pregnancy and is where the premature funneling starts to happen. To characterize the cervical collagen ultrastructure for the upper half of the human cervix, we imaged whole axial tissue slices from non-pregnant and pregnant women undergoing hysterectomy or cesarean hysterectomy respectively using optical coherence tomography (OCT) and implemented a pixel-wise fiber orientation tracking method to measure the distribution of fiber orientation. The collagen fiber orientation maps show that there are two radial zones and the preferential fiber direction is circumferential in a dominant outer radial zone. The OCT data also reveal that there are two anatomic regions with distinct fiber orientation and dispersion properties. These regions are labeled: Region 1-the posterior and anterior quadrants in the outer radial zone and Region 2-the left and right quadrants in the outer radial zone and all quadrants in the inner radial zone. When comparing samples from nulliparous vs multiparous women, no differences in these fiber properties were noted. Pregnant tissue samples exhibit an overall higher fiber dispersion and more heterogeneous fiber properties within the sample than non-pregnant tissue. Collectively, these OCT data suggest that collagen fiber dispersion and directionality may play a role in cervical remodeling during pregnancy, where distinct remodeling properties exist according to anatomical quadrant.
Pub.: 30 Nov '16, Pinned: 31 Aug '17
Abstract: Abnormal changes in orientation of myofibers are associated with various cardiac diseases such as arrhythmia, irregular contraction, and cardiomyopathy. To extract fiber information, we present a method of quantifying fiber orientation and reconstructing three-dimensional tractography of myofibers using optical coherence tomography (OCT). A gradient based algorithm was developed to quantify fiber orientation in three dimensions and particle filtering technique was employed to track myofibers. Prior to image processing, three-dimensional image data set were acquired from all cardiac chambers and ventricular septum of swine hearts using OCT system without optical clearing. The algorithm was validated through rotation test and comparison with manual measurements. The experimental results demonstrate that we are able to visualize three-dimensional fiber tractography in myocardium tissues.
Pub.: 25 Oct '13, Pinned: 31 Aug '17
Abstract: Optical coherence tomography (OCT) is able to provide high resolution volumetric data for biological tissues. However, the field of view (FOV) of OCT is sometimes smaller than the field of interest, which limits the clinical application of OCT. One way to overcome the drawback is to stitch multiple 3D volumes. In this paper, we propose a novel method to register multiple overlapped volumetric OCT data into a single volume. The relative positions of overlapped volumes were estimated on en face plane and at depth. On en face plane, scale invariant feature transform (SIFT) was implemented to extract the keypoints in each volume. Based on the invariant features, volumes were paired through keypoint matching. Then, we formulated the relationship between paired offsets and absolute positions as a linear model and estimated the centroid of each volume using least square method. Moreover, we calibrated the depth displacement in each paired volume and aligned the z coordinates of volumes globally. The algorithm was validated through stitching multiple volumetric OCT datasets of human cervix tissue and of swine heart. The experimental results demonstrated that our method is capable of visualizing biological samples over a wider FOV, which enhances the investigation of tissue structure such as fiber orientation.
Pub.: 09 Jan '15, Pinned: 31 Aug '17
Abstract: During pregnancy, the uterine cervix is the mechanical barrier that prevents delivery of a fetus. The underlying cervical collagen ultrastructure, which influences the overall mechanical properties of the cervix, plays a role in maintaining a successful pregnancy until term. Yet, not much is known about this collagen ultrastructure in pregnant and nonpregnant human tissue. We used optical coherence tomography to investigate the directionality and dispersion of collagen fiber bundles in the human cervix. An image analysis tool has been developed, combining a stitching method with a fiber orientation measurement, to study axially sliced cervix samples. This tool was used to analyze the ultrastructure of ex-vivo pregnant and non-pregnant hysterectomy tissue samples taken at the internal os, which is the region of the cervix adjacent to the uterus. With this tool, directionality maps of collagen fiber bundles and dispersion of collagen fiber orientation were analyzed. It was found that that the overall preferred directionality of the collagen fibers for both the nonpregnant and pregnant samples were circling around the inner cervical canal. Pregnant samples showed greater dispersion than non-pregnant samples. Lastly, we observed regional differences in collagen fiber dispersion. Fibers closer to the inner canal showed more dispersion than the fibers on the radial edges.
Pub.: 25 Apr '15, Pinned: 31 Aug '17
Abstract: We present an ultrahigh-resolution spectral domain optical coherence tomography (OCT) system in 800 nm with a low-noise supercontinuum source (SC) optimized for myocardial imaging. The system was demonstrated to have an axial resolution of 2.72 μm with a large imaging depth of 1.78 mm and a 6-dB falloff range of 0.89 mm. The lateral resolution (5.52 μm) was compromised to enhance the image penetration required for myocardial imaging. The noise of the SC source was analyzed extensively and an imaging protocol was proposed for SC-based OCT imaging with appreciable contrast. Three-dimensional datasets were acquired ex vivo on the endocardium side of tissue specimens from different chambers of fresh human and swine hearts. With the increased resolution and contrast, features such as elastic fibers, Purkinje fibers, and collagen fiber bundles were observed. The correlation between the structural information revealed in the OCT images and tissue pathology was discussed as well.
Pub.: 24 Mar '16, Pinned: 31 Aug '17