publications

2023

1. Segmentation

   

   Automatic Multiple Articulator Segmentation in Dynamic Speech MRI Using a Protocol Adaptive Stacked Transfer Learning U-NET Model

   Bioengineering, 2023

   Abs HTML

   Dynamic magnetic resonance imaging (MRI) is a valuable technique for studying upper-airway function during speech production by analyzing changes in vocal tract airspace and the positioning of soft-tissue articulators. This paper introduces a stacked transfer learning U-NET model aimed at segmenting the deforming vocal tract in 2D mid-sagittal slices of dynamic speech MRI. The model utilizes both low- and mid-level features from pre-trained datasets (brain tumor MR, lung CT, and an in-house airway dataset) and high-level features from protocol-specific MR images. Demonstrated across three fast speech MRI protocols with varying acquisition schemes and speech tasks, the method achieved accurate segmentations comparable to expert human annotators, even with limited protocol-specific images. Evaluation metrics included the DICE similarity coefficient, Hausdorff distance, and segmentation count, affirming the model’s effectiveness and adaptability across different MRI protocols.

2022

1. Reconstruction

   

   Accelerated Pseudo 3D Dynamic Speech MR Imaging at 3T Using Unsupervised Deep Variational Manifold Learning

   MICCAI, 2022

   Abs HTML

   This work presents a novel approach using unsupervised deep variational manifold learning to enhance magnetic resonance imaging (MRI) of the vocal tract during speech. By recovering a "pseudo-3D" dynamic speech dataset from sequential 2D slices, the method achieves high temporal resolution (18 ms) while addressing the limitations of current volumetric MR techniques, which struggle with spatio-temporal resolution, organ coverage, and signal-to-noise ratio. The approach utilizes a 3D convolutional neural network to create volumes of the vocal tract, enforcing conditions for smoothness, regularization, and data consistency. Evaluated using in-vivo data from normal volunteers, this method allows for the first-time extraction of quantitative 3D vocal tract area functions from under-sampled 2D datasets, aiding in the understanding of vocal tract shape changes during speech

2020

1. Segmentation

   

   Airway segmentation in speech MRI using the U-net architecture

   Sajan Goud Lingala Subin Erattakulangara

   International Symposium on Biomedical Imaging (ISBI), 2020

   Abs HTML

   We develop a fully automated airway segmentation method to segment the vocal tract airway from surrounding soft tissue in speech MRI. We train a U-net architecture to learn the end to end mapping between a mid-sagittal image (at the input), and the manually segmented airway (at the output). We base our training on the open source University of Southern California’s (USC) speech morphology MRI database consisting of speakers producing a variety of sustained vowel and consonant sounds. Once trained, our model performs fast airway segmentations on unseen images at the order of 210 ms/slice on a modern CPU with 12 cores. Using manual segmentation as a reference, we evaluate the performances of the proposed U-net airway segmentation, against existing seed-growing segmentation, and manual segmentation from a different user.

2019

1. Genetics

   

   Network Analysis of MPO and Other Relevant Proteins Involved in Diabetic Foot Ulcer and Other Diabetic Complications

   T. V. Suchithra Mathew Saumya

   Interdisciplinary Sciences: Computational Life Sciences, 2019

   Abs HTML

   Network analysis and visualization of genes are very important to understand large complex biological data in a better manner. Large data on genes and proteins in the biological systems are analyzed on the occurrence, interactions, co-expression, and co-regulations of various genes. Here we have visualized the genes involved in type 1 diabetes (T1D), type 2 diabetes (T2D), and foot ulcer condition to put light on the corrective measures to the problem of impaired healing. The goal of this study was to identify the important genes involved in the pathogenesis of diabetes complications and foot ulcer and its association with the free radical-producing enzyme, the myeloperoxidase (MPO). In this study, we have used bioinformatics tools for the analysis of 24 genes that play a major role in diabetes mellitus and its complications, especially diabetic foot ulcer to reveal the relation between the genes and proteins involved in these disease conditions. We could conclude from the network model that MPO is related to foot ulcer and involved in pathogenesis of various co-associated diseases, such as oxidative stress, inflammation, peripheral vascular disease, and other related diabetes complications