Projects
Quick Navigation Links:
Respiration-induced Tumor Motion Analysis
Collaborator: Department of Radiation Oncology (Medical Physics Division), UT Southwestern Medical Center, Dallas.
This research project is aimed at enhancing the efficiency of radiation therapy for treating abdominal and thoracic tumors. The objective is to analyze tumor motion traces and accompanying surface motion to:
adaptive data processing strategy tailored to the patient’s respiratory characteristics
algorithms for real-time prediction of irregularities and anomalies
efficient techniques and tools for Predictive Modeling and Association Rule Mining in 4DCT data and VisionRT based surface motion tracking of the chest and abdominal regions
temporal classification of breathing types based on the variation in the motion patterns
explore pattern distributions and similarity based patient grouping to facilitate better treatment planning and personalization
Read More:
A. Balasubramanian, R. Shamsuddin, B. Prabhakaran, A. Sawant. Predictive modeling of respiratory tumor motion for real-time prediction of baseline shifts. Physics in Medicine & Biology, 2017.
R. Shamsuddin, A. Balasubramanian, A. Sawant and B. Prabhakaran. Calculating Patient Similarity Based on Respiration Induced Tumor Motion. IEEE International Conference on Healthcare Informatics 2015 (ICHI 2015), Dallas, USA, October 2015.
A. Balasubramanian, R. Shamsuddin, Y. Cheung, A. Sawant and B. Prabhakaran. Exploring Baseline Shift Prediction in Respiration Induced Tumor Motion. IEEE International Conference on Healthcare Informatics 2014 (ICHI 2014), Verona, Italy, September 2014.
A. Balasubramanian, D. Kim, Y. Cheung, A. Sawant and B. Prabhakaran. Analysis of Surface Motion Patterns Changes for Detecting Baseline Shifts in Respiratory Tumor Motion Data. 3rd Workshop on Data Mining for Medicine and Healthcare (DMMH), 14th SIAM International Conference on Data Mining (SDM 2014), Philadelphia, USA, April 2014.
A. Balasubramanian, B. Prabhakaran, and A. Sawant. Mining Pattern Sequences in Respiratory Tumor Motion Data. International Conference of Engineering in Medicine and Biology Society (EMBC 2012), San Diego, USA, August 28–September 1, 2012.
VisibleSpeech (Opti-Speech)
Project Website: http://visiblespeech.utdallas.edu
Collaborators: Callier Center for Communication Disorders, UT Dallas, Vulintus Inc., Dallas, Northern Digital Inc., Canada, Carstens Medizinelektronik GmbH, Germany.
VisibleSpeech (or Opti-Speech) is a framework for real-time capture and visualization of human tongue motion, to assist clinical diagnosis and rehabilitation of patients with speech disorders, such as in case of Dysarthria and ALS (Lou Gehrig's disease).
Features:
a virtual tongue model driven in real-time by tongue motion capture data, with offline playback functionalities.
a database of tongue motions and speech production data, and associated toolkit to facilitate inter-disciplinary research
allow the clinicians to visualize and compare the motion characteristics of the tongue, jaw and lips longitudinally, within patients as well as within and between various patient groups, along with corresponding audio information (optional)
provide visual feedback to patients during rehabilitative training tasks.
I am involved in developing techniques to identify patterns and pattern associations in motion trajectories and automatic classification of speech pronunciations in the absence of audio data.
Techniques: Hidden Markov Models, Bayesian Modeling, Support Vector Machines and Association Rule Mining (C++, C#, Java, MATLAB, Autodesk MotionBuilder, Unity Game Engine).
Read More:
W. Katz, T. Campbell, J. Wang, E. Farrar, J.C. Eubanks, A. Balasubramanian, L. Mojica, B. Prabhakaran and R. Rennaker. OptiSpeech: A real-time, 3D visual feedback system for speech training. 15th Annual Conference of International Speech Communication Association (InterSpeech 2014), Lyon, France, September 2013.
J. Wang, A. Balasubramanian, L. Mojica, J. Green, A. Samal and B. Prabhakaran. Word Recognition from Continuous Articulatory Movement Time-series Data using Symbolic Representations. 4th Workshop on Speech and Language Processing for Assistive Technologies (SLPAT), Annual Conference of International Speech Communication Association (InterSpeech 2013), Lyon, France, August 2013.
Human Activity Recognition using Smartphone-based Sensors
In this project, we explored the recognition of human activities (walk, run, jog, climb stairs up, climb stairs down, stand still) from data acquired using smartphone-based inertial sensors -- accelerometer, gyroscope and magnetometer.
Features:
detecting activities irrespective of the user specific mobile sensor position, orientation and body attachment
compare the classification accuracies in case of orientation corrected and uncorrected data synchronously captured from smartphones in different positions and and orientations across multiple individuals
Techniques: Logistic Regression, Support Vector Machines, Bagging (MATLAB, Objective C, C++, Java).
Read More:
L. Mojica, S. Raghuraman, A. Balasubramanian and B. Prabhakaran. Exploring Unconstrained Mobile Sensor Based Human Activity Recognition. 3rd International Work- shop on Mobile Sensing, ACM/IEEE Information Processing in Sensor Networks (IPSN 2013), Philadelphia, USA, April 2013.
PatientDoc Health Record Indexing
I am currently working on developing a robust record index for storage and retrieval of multidimensional biomedical sensor data. The features of this index are:
compressed multidimensional time series data from heterogenous sensor arrays, and summary of intrinsic characteristics and semantic annotations
techniques for efficient discovery of new multidimensional patterns, anomalies and regions of interest, with support for automatic identification of known patterns and content based query resolution.
Techniques: Unsupervised and semi-supervised modeling, Structural similarity modeling, String encoding based discretization, Dynamic Time Warping, Multi-resolution modeling using trie-based structures, Association Rule Mining (MATLAB, C++, Java, Python).
Read More:
A. Balasubramanian, J. Wang and B. Prabhakaran. Discovering Multidimensional Motifs in Physiological Signals for Personalized Healthcare. IEEE Journal of Selected Topics in Signal Processing, 2016.
A. Balasubramanian and B. Prabhakaran. Flexible Exploration and Visualization of Motifs in Biomedical Sensor Data. Workshop on Data Mining for Healthcare (DMH), ACM Conference of Knowledge Discovery and Data Mining (KDD 2103), Chicago, USA, August 2013.
A. Balasubramanian, J. Wang and B. Prabhakaran. Multidimensional Motif Discovery in Physiological and Biomedical Time Series Data. Technical Report UTDCS-03-14, Department of Computer Science, University of Texas at Dallas, February 2014.