I'm Saket, a Biomedical Engineer and Deep learning enthusiast. I'm completing my Master's degree at The University of Texas at Dallas and working as a researcher at UT Southwestern. Prior to this, I obtained my undergraduate and Masters' degrees in Biological Engineering from the Indian Institute of Technology Madras.
At UT Southwestern, I work in the Deep Learning for Precision Health Lab with Dr. Albert Montillo along with Dr. Ted Mau.
My Projects
At UTSW Medical Center:
LIVER FIBROSIS
Collaborators: Dr. Albert Montillo, Deep Learning for Precision Health LabEstimating liver fibrosis severity from Ultrasound texture
Methods: Deep Learning, Ultrasound Image processing
I adopted a deep learning approach (fully convolutional neural network) within an image processing pipeline for despeckling, smoothing and segmenting, to analyse a dataset of liver fibrosis patients at the UTSW medical center, for assessing the extent of granularity in the ultrasound images as a surrogate for tissue punch biopsies- a follow up project from the previous year’s UHackMed codeathon.
ASSOCIATIVE MEMORY
Collaborators: Dr. Brad Lega , Texas Computational Memory LabClassification of intracranial EEG signals base don Hippocampal recordings
Methods: Statistical Machine Learning, Signal processing
During this year’s UHackMed codeathon, I built a cepstrum based machine learning pipeline to classify episodic memory events from intracranial iEEG recordings, as strong (successful association) vs weak (familiar items). The pipeline successfully (76% accuracy) accomplished subject independency utilizing only hippocampal electrodes while reducing the signal duration for achieving accurate classification from 700ms to 600ms.
SPEECH REHAB
Collaborators: Dr. Ted Mau , Dept of OtolaryngologyQuantifying vestibular fold viscoelasticity and vocal quality from laryngoscope videography
Methods: Statistical Clustering, Signal processing, Deep Learning, Computer vision
I am building automated surgical decision support pipelines to aid in forecasting tissue response during voice rehab. I am adapting cutting edge computer vision algorithms originally developed for behavioral neuroscience, to assess vestibular fold movement as a potential surrogate for predicting post laryngectomy voice outcome.
At UTD:
CARDIOVASCULAR IMAGING
Collaborators: Dr. Kenneth Hoyt , Ultrasound Imaging and Therapy LaboratoryReal-time extraction of carotid artery lumen diameter from smartphone-powered Ultrasound imaging
Methods: Ultrasound Image processing
I adopted a string of frequency domain ML algorithms for converting raw RF ultrasound images to B-mode and, despeckling, smoothing and segmenting them in order to visually enhance arterial wall boundary locations, for automated accurate extraction of inner and outer diameters based on the laplacian of elastography-intensity.
NEURAL CORRELATES OF SPEECH
Collaborators: Dr. Jun Wang , Speech Disorders and Technology LaboratoryDecoding acoustic speech representation and articulator motion from MEG signals
Methods: Deep learning, Statistical Machine learning, Signal processing
We tested automated time-series ML pipelines to extract mapping between individual brain waves and speech patterns, to successfully forecast jaw motion and vocal onset and endset, purely from the analysis of MEG signals of subjects performing a speech production task.
DEMAND FORECASTING
Collaborators: Dr. Gautam Kunapuli , Statistical Relational Learning LaboratoryForecasting hourly load in bike-sharing systems from time-series weather data
Methods: Deep learning, Statistical Machine learning, Signal processing
We tested automated time-series ML pipelines to extract mapping between individual brain waves and speech patterns, to successfully forecast jaw motion and vocal onset and endset, purely from the analysis of MEG signals of subjects performing a speech production task.
At IITM:
HYDROGEL SCAFFOLDS
Collaborators: Jayakrishnan A , Dept. of BiotechnologyNovel In situ forming adhesive hydrogels for wound management applications
Methods: Scanning Electron Microscopy (SEM), Rheology
I characterized swelling, morphology (SEM), viscoelasticity (rheology), drug encapsulation and release profiles of polyaldose-gels