Recent Key efforts from The Innovation Center For Biomedical Informatics (ICBI)
- The Innovation Center For Biomedical Informatics (ICBI) at Georgetown University has been collaborating with the US Dept. of Veterans Affairs (VA) and the Georgetown Department of Psychiatry to develop innovative e-Mental Health interventions that leverage our expertise in mHealth and telemedicine. Most recently, we mined acoustic and semantic features from audio interviews to predict suicidal tendencies in military veterans. Using the 208 narrative audios collected from veterans, a classifier was built that differentiates suicidal from non-suicidal veterans based on acoustic features of speech and sentiment analysis of the transcribed narratives. The classifier correctly identified veterans with suicidal ideation from with an overall accuracy of 86.2% and area under the receiver operating characteristic curve (AUC) equal to 83.6%. This work, done using tools such as Google speech-to-text and Natural Language Processing (NLP) APIs and Watson Tone Analyzer, was presented in a poster session at the Technology in Psychiatry Summit 2018. Correlating different types of data about the veterans will help identify veterans at higher risk of suicide in a clinical setting. Figure 1 shows the workflow diagram for this project.
Figure 1: Study Workflow For the Suicide Ideation Detection Project
- The ICBI team participated in the 2018 Data Science Bowl. The Data Science Bowl is a worldwide competition that brings together data scientists, technologists, and domain experts across industries to take on the world’s challenges with data and technology. In the 2018 Data Science Bowl challenge, the aim was to identify the nuclei in divergent microscopy images, regardless of the experimental setup, over a period of 90 days. Our team approached this automated identification challenge in three different ways and scored in the top 12% out of more than 68,000 algorithms that were submitted. Figure 2 shows an example of automated recognition of the nuclei from pathology images using a deep learning algorithm.
Figure 2: The image shows the automated recognition of the nuclei from pathology images using a deep learning algorithm
- We are currently working on integrating molecular data with information extracted from radiology MRI images from a large collection of brain cancer patients, called the REMBRANDT study. The goal of this computational analysis is to better understand how immune cells affect clinical outcome and survival in brain cancer patients, and ultimately improving treatment options.
- In the past, we have used machine learning Support Vector Method (SVM) algorithms to integrate genome wide multi-modal molecular data from cancer patients. Gene expression, microRNA, copy number and metabolomics data from urine and serum were integrated to identify biomarkers strongly associated with recurrence (Madhavan et al, Frontiers in Genetics, 2013). Figure 3 shows the Bioinformatics workflow of this multivariate analysis
Figure 3: Bioinformatics workflow of multivariate analysis
Artificial Intelligence efforts from our collaborators at Georgetown University
- The Information Retrieval Lab, part of the Georgetown University's Computer Science Department has built an AI tool that is currently in use at Georgetown University Radiology Department to detect the discrepancies in the radiology reports. This work has been published at the Workshop on data Mining for Medicine and Healthcare at SDM 2016, and also has a patent application.
- The Information Retrieval Lab employed neural networks to predict depression and self-harm severity based on text in online help forums, allowing moderators to give attention to those whose lives might be at risk. This work received the “Best Long Paper” award at the EMNLP 2017 conference (Yates et al, EMLNP 2017). The work was later expanded to include 9 mental health conditions which received Area Chair Favorite (contender for best paper award) at COLING 201 (Cohan et al, JASIST 2017).
- The Information Retrieval Lab also applied a convolutional neural model aimed at improving clinical notes. Clinical notes are part of a medical record where healthcare professionals record details about a patient's progress. Denoising these clinical notes making them more suitable for retrieval of relevant medical literature (Soldaini et al, CIKM 2017 ).
- The Information Retrieval Lab developed a neural network architecture for categorizing Patient Safety Events. Patient Safety Event reports are narratives describing potential adverse events to the patients and are important in identifying and preventing medical errors (Cohan et al, European Conference on Information Retrieval, 2017). Similarly, we used attentive convolutional and recurrent networks for identifying harm events in patient care and categorize the harm based on its severity level (Cohan et al, ACM-BCB, 2017)
- The Computer Science Department, in collaboration with the Massive Data Institute (MDI) at Georgetown University, used a mix of graph-based topic modeling and event detection algorithms to understand the types of content shared, and SVM and logistical regression models for demographic inference of gender and age to mine the #metoo movement in Twitter, which is a global movement against sexual harassment and assault, particularly in the workplace. These algorithms were also been applied on Twitter conversations related to election dynamics during the 2016 presidential election (Bode et al.; to appear and extremism/terrorism (Wei et al. PAKDD 2017).
- Extracting Adverse Drug Reactions from Social Media:
- A. Yates and N. Goharian, "ADRTrace: Detecting Expected and Unexpected Adverse Drug Reactions from User Reviews on Social Media Sites", In Proceedings of the 35th European Conference on Information Retrieval (ECIR 2013), 2013.
- A. Yates, N. Goharian, O. Frieder, "Extracting Adverse Drug Reactions from Social Media", Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence (AAAI-15), 2015.
- Public Health Surveillance using Twitter: J. Parker, Y. Wei, A. Yates, O.Frieder, and N. Goharian, "A Framework for Detecting Public Health Trends with Twitter", The 2013 IEEE/ACM International Conference on Advances in Social Network Analysis and Mining, Aug. 2013