Research‎ > ‎

Medical Image/Signal Processing

While I am interested in many computational problems in medicine, I am most fascinated by Computational Neuroscience, a discipline where algorithms and software tools meet neuroscience. I am particularly interested in brain imaging (e.g., MRI) and brain electromagnetic signals (e.g., EEG).


A quote on the wall of Montreal Neurological Institute, McGill University

Epilepsy Diagnosis Using Scalp Interictal EEG

Conventional epilepsy diagnosis based on EEG requires the occurrenc of seizures or interictal epileptiform discharges (IDEs), such as slow waves and spikes. As a result, long-term EEG monitoring, often with joint video recording, is needed. However, this solution is both tedious and costly.

Since a patient is in the interictal state (i.e., the time between seizures) for most of the time, it would be very useful if epilepsy diagnosis can be done by using short-term (e.g., 20 minutes) scalp interictal EEG.

Because epilepsy is caused by the dysfunction of the brain, normal EEG and epileptic EEG (even during interictal periods) may have different "fingerprints". These differences are  probably too subtle for the human eye to see, but it could be detectable using proper computational tools.

My approach extracts features from EEG signals (i.e., quantifying each EEG segment into an array of numeric values) and then runs classifiers (e.g., SVMs) on these features to discriminate between interictal epileptic EEG and normal EEG, and thus distinguish epilepsy patients from healthy people.

A pilot study was done from 2008 to 2010, delivering promising results from two datasets (one of scalp recordings with artifacts and the other of intrancranial recordings without artifacts). One of the papers of the pilot study was reported on MIT Technology Review Website. The other one was reported by 2009's special research issue (PDF) of epilepsyUSA, the official magazine of Epilepsy Foundation of America.

Please note that this work is different from seizure detection and spike detection, both of which still require long-term EEG recording. Nor is this work the same as seizure prediction.

In addition to this open question, I am also interested in any other EEG/MEG processing topics, including but not limited to, focus localization, portable/wearable devices for epilepsy care giving, seizure prediction, spike detection.

Collaborators

  • Justin Dauwels, when he was doing postdoc at MIT and MGH, independently and concurrently developed a similar idea. We met at IEEE EMBC 2009 and were both shocked how close our ideas were. He focuses on using synchrony measures to study epileptic EEG. Now he is with Nanyang Technical University, Singapore.
  • Lunal Khuon, director of Biomedical Electronics Lab at Villanova, built an Android-based seizure detection system from multi-modal physiological signals.

MindBoggle: MRI Morphometry and Mental Disorders

I have been very fortunate to work with a great team on a project called MindBoggle (at GitHub). The project was initialized by Dr. Arno Klein via a 3-year NIH R01 grant at Columbia Medical School, and involves people from MIT/Harvard, Stony Brook University (SUNY Stony Brook), U. Washington, and Université Catholique de Louvain in Belgium. (e.g., our long authors and affiliations list)

MindBoggle aims to extract anatomical landmarks, e.g., fundi at the bottom of sulci,  from human brain cortex and to compare morphological features of these landmarks across brains. Previous studies have shown that morphological changes on these landmarks can be related to Alzheimer's Disease, depression (our SfN 2011 presentation), etc. My main contribution to MindBoggle is landmark extraction and characterization. 

   

Medical Instrumentation

In terms of medical instrumentation, my experience is on low-cost physiological signal (e.g., ECG, EEG) sampling circuit and smartphone-based medical systems. Please check Electronics page for more information.


I designed an ECG sampling circuit using two Texas Instruments INA128's to amplify the signal by 80dB (10000 times) and a National Instruments myDAQ to pass the signal to LabVIEW. The ECG signal quality is very good while the cost is very low (<$10 for one INA128).

I was involved in a research led by Lunal Khuon on building a seizure detection solution using multi-modal physiological signals and an Android phone. The signals are sampled using low-cost circuits.

MEG

I work on MEG from time to time. In the summer of 2011, I was involved in a project that registered SAM results from MEG with MRI images. Recently I am working on analyzing MEG data from surgery-verified epilepsy patients and healthy people. The similarity between MEG and EEG allows me to port algorithms for EEG onto processing MEG data.

Collaboration opportunity

  • In order to tes my approaches in epileptic EEG research (e.g., epileptic interictal scalp EEG recognition, smartphone-based seizure detection, etc.), I am in great need of a large-scale scalp EEG dataset (of at least 100 epileptic subjects and matching normal people). If you have such a dataset, or can help me build such a dataset, please contact me.
  • I am also open to any medical data analysis or medical instrumentation/electronics projects.