Advertisement

AI and Big Data in Psychiatry: An Introduction and Overview

Psychiatrist, Data Scientist
Apr. 15, 2019
Advertisement

More Related Content

Slideshows for you(20)

Similar to AI and Big Data in Psychiatry: An Introduction and Overview(20)

Advertisement

More from Carlo Carandang(20)

Advertisement

AI and Big Data in Psychiatry: An Introduction and Overview

  1. Big Data in Psychiatry: An Introduction and Overview Carlo Carandang, MD, MSc, FAPA
  2. Learning Goals • Understand Big Data in Psychiatry • Understand Big Data Systems for Psychiatry • Understand machine learning algorithms for Psychiatry AI applications
  3. I, Robot (2004) • Det. Spooner : So, Dr. Calvin, what exactly do you do around here? • Dr. Calvin : My general fields are Advanced Robotics and Psychiatry. Although, I specialize in hardware-to-wetware interfaces in an effort to advance U.S.R.'s robotic ahthropomorphization program. • Det. Spooner : So, what exactly do you do around here? • Dr. Calvin : I make the robots seem more human. • Det. Spooner : Now wasn't that easier to say? • Dr. Calvin : Not really. No. This Photo by Unknown Author is licensed under CC BY
  4. World’s 1st Mental Health Chatbot (2018) This Photo by Unknown Author is licensed under CC BY
  5. Definitions • AI aims to mimic human cognitive functions • AI is built from: • Machine Learning for structured data • Natural Language Processing (NLP) for unstructured data • Big Data • petabytes (1,024 terabytes): millions of patient records • Hard Coding • explicit coding instructions to carry out a task • Machine Learning • computer use of data, algorithms, and statistical models to perform a task without being hard-coded • relies on patterns and inference • Linear algebra: basis for machine learning • Scalar: magnitude x • Vector: magnitude and direction (x,y) • Matrix: array of elements in row and columns This Photo by Unknown Author is licensed under CC BY-SA-NC This Photo by Unknown Author is licensed under CC BY-SA This Photo by Unknown Author is licensed under CC BY-SA-NC
  6. The road map from clinical data generation to natural language processing data enrichment, to machine learning data analysis, to clinical decision making. EMR, electronic medical record; EP, electrophysiological. Fei Jiang et al. Stroke Vasc Neurol 2017;2:230-243 © Article author(s) (or their employer(s) unless otherwise stated in the text of the article) 2017. All rights reserved. No commercial use is permitted unless otherwise expressly granted.
  7. Publications of Machine Learning in Psychiatry Reevaluating the Efficacy and Predictability of Antidepressant Treatments: A Symptom Clustering Approach. JAMA Psychiatry 2017 Multisite Prediction of 4-Week and 52-Week Treatment Outcomes in Patients With First-Episode Psychosis: A Machine Learning Approach. Lancet Psychiatry 2016 Predicting Response to Repetitive Transcranial Magnetic Stimulation in Patients With Schizophrenia Using Structural Magnetic Resonance Imaging: A Multisite Machine Learning Analysis. Schizophrenia Bulletin 2018 Treatment Response Prediction and Individualized Identification of First-Episode Drug-Naïve Schizophrenia Using Brain Functional Connectivity. Molecular Psychiatry 2018
  8. Use Big Data, Systems, and Algorithms to Solve Psychiatry Problems Big Data Big Data Systems Algorithms Psychiatry AI Applications This Photo by Unknown Author is licensed under CC BY-SA
  9. Use Big Data, Systems, and Algorithms to Solve Psychiatry Problems Big Data Big Data Systems Algorithms Psychiatry AI Applications This Photo by Unknown Author is licensed under CC BY-SA
  10. This Photo by Unknown Author is licensed under CC BY-SA
  11. • $300 billion potential annual value to Healthcare • $165 billion from clinical decision support • $108 billion from personalized medicine and clinical trial design • $47 billion from fraud detection • $9 billion from public health surveillance and response • $5 billion from digitization, aggregation, and automation of patient records • 30% of data produced worldwide is from healthcare This Photo by Unknown Author is licensed under CC BY
  12. From Data to Wisdom This Photo by Unknown Author is licensed under CC BY
  13. Psychiatry generates lots of different kinds of information • Clinical history, patient demographics, diagnosis, procedures (psychotherapy/ECT/TMS), medication, lab results, and clinical notes. • On-body sensors and other devices that patients wear (activity, behavior), social media (thoughts, mood, reactions of social network to patient). • Real-time data sources such as blood pressure, temperature, heart rate (sympathetic nervous system and connection to Amydala-driven fear response). • Drug dispensing levels at acute psychiatric inpatient units. This Photo by Unknown Author is licensed under CC BY-SA
  14. Use Big Data, Systems, and Algorithms to Solve Psychiatry Problems Big Data Big Data Systems Algorithms Psychiatry AI Applications This Photo by Unknown Author is licensed under CC BY-SA
  15. Hadoop is the platform that administers MapReduce to process and analyze big data Patient 1 Patient 2 Map Map MDD, 1 GAD, 1 PTSD, 1 MDD, 1 DATA
  16. Hadoop is the platform that administers MapReduce to process and analyze big data MDD, 1 GAD, 1 PTSD, 1 MDD, 1 MDD: 1, 1 GAD: 1 PTSD: 1 Reduce MDD: 2 GAD: 1 PTSD: 1 MapReduce in Hadoop
  17. Pros/Cons of MapReduce in Hadoop • Pro: Hadoop can work with big data using MapReduce computations • Con: Hadoop not good for iterative computations, such as required for machine learning • Pro: Hadoop works well for billions of records that you want a histogram, odds ratio, or prevalence • Pro and Con: Hadoop uses distributed hard discs to store massive data, but also makes it slow for iterative computations This Photo by Unknown Author is licensed under CC BY-SA
  18. Spark is the platform for machine learning of Big Data • Spark uses RAM, so it’s faster than Hadoop MapReduce data stored on magnetic disks • Spark is fast enough to perform iterative computations needed by machine learning • Spark uses distributed RAM to handle massive computations of Big Data This Photo by Unknown Author is licensed under CC BY-SA
  19. Use Big Data, Systems, and Algorithms to Solve Psychiatry Problems Big Data Big Data Systems Algorithms Psychiatry AI Applications This Photo by Unknown Author is licensed under CC BY-SA
  20. Supervised versus Unsupervised Learning
  21. Graphical illustration of unsupervised learning, supervised learning and semisupervised learning. Fei Jiang et al. Stroke Vasc Neurol 2017;2:230-243 © Article author(s) (or their employer(s) unless otherwise stated in the text of the article) 2017. All rights reserved. No commercial use is permitted unless otherwise expressly granted.
  22. Classification Classification algorithm takes that input x which is in matrix. And for example every row in this matrix represents a patient and every column of this matrix represent a feature. And the classification algorithm will map this input feature matrix x into a set of discrete outcome y. For example, these y vectors indicate whether the patient had Treatment-Resistant Depression or not. So we want to learn this function f that mapped input x to the target outcome variable y, this is classification. y = f(x) + b feature x patient f outcome y
  23. Regression Regression algorithm takes that input x which is in matrix. And for example every row in this matrix represents a patient and every column of this matrix represent a feature. And the classification algorithm will map this input feature matrix x into a set of continuous outcome Y. For example, these Y vectors indicate the GAF score of the patient. So we want to learn this function f that mapped input x to the target outcome variable Y, this is regression. outcomefeature yx patient y = f(x) + b f
  24. Clustering Clustering is about taking an input matrix x, again, every row here is a Patient (P), every column is a feature, Disease (D). Want to learn a function, f1, that partitioned this matrix into multiple clusters, P1, P2, and P3, and f2 to D1, D2, and D3. In f1, each cluster consists of set of patients and those patients are similar to each other, and those patients within a cluster will be different patients in a different cluster. f2 into different disease phenotypes. Patient (P) Disease (D) P2 P3 P1 x Patient stratification D1 D2 D3 Computational Phenotyping
  25. Measuring Performance: Classification This Photo by Unknown Author is licensed under CC BY
  26. Measuring Performance: Regression • 1 indicates regression line perfectly fits data • 0 indicates regression line does not fit data at all R squared (coefficient of determination) • Error between prediction and actual value • Small for low error, large for high error Mean Absolute Error (MAE), Mean Squared Error (MSE)
  27. Measuring Performance: Clustering • Rand Index (RI) • 0 means bad clustering assignment • 1 means perfect clustering assignment
  28. Use Big Data, Systems, and Algorithms to Solve Psychiatry Problems Big Data Big Data Systems Algorithms Psychiatry AI Applications This Photo by Unknown Author is licensed under CC BY-SA
  29. Predictive Modeling Predictive modeling is about using historical data to build algorithms to predict future events. For example, we can predict which treatment is likely to work for a patient with depression. Example: Let’s look at patients with depression who responded to treatment. Group A: responded in first two years of treatment. Group B: responded between two to five years of treatment. Group C: continued to suffer even after five years of treatment. Predictive modeling would help improve matching patients to the right treatment quickly, so that the late responders in group B will become early responders in group A. Also, this will help identify non-responders in group C quickly, so that a new treatment can be developed for them. Group A Group B Group C 35% 20% 45% Early Responders Predictive Modeling: Match to right Rx quickly New Rx Predictive Modeling: Match to new Rx quickly
  30. Computational Phenotyping The input to Computational Phenotyping is the raw patient data. It consist of many different sources, such as demographic information, diagnosis, medication, procedure, lab tests, and clinical notes. And phenotyping is the process of turning the raw patient data into medical concepts or phenotypes. Phenotyping Phenotypes Demographic Diagnosis Medication Procedure Psychotherapy Clinical Notes Lab tests Psychological Testing Raw Data
  31. Patient Similarity When treating patients, doctors often compare the current patient to past patients they have seen, also known as case-based reasoning. Patient similarity is about simulating the doctor's case-based reasoning using computer algorithms. Instead of depending on one doctor's memory, wouldn't it be nice if we can leverage all the patient data in the entire database? So when the patient comes in, the doctor does an examination of the patient. Then, based on that information, we can do a similarity search through the database. Find those potentially similar patients, then the doctor can provide some supervision on that result to find those truly similar patients through this specific clinical context. Then, we can group those patients based on what treatment they are taking and look at what outcome they are getting. Then, recommend the treatment with the best outcome to the current patient. Patient Doctor Patient Database
  32. How to Use Patient Similarity in Clinical Practice Guideline available? Similar patients? Ready for RCT? Use Patient Similarity Use clinical judgment Develop hypotheses Perform RCT Use guideline Yes No No Yes Yes No
  33. ‘Perils’ of Machine Learning in Psychiatry • “There have been so many fads, and we’ve discovered the ultimate answer in psychiatry a thousand times,” Rajiv Tandon said in an interview. “Machine learning could be just another one.” • The biggest challenges are sensitivity and specificity of an algorithm’s predictive results. Ultimately, the results of any algorithm must be replicable across different datasets. “This is a particularly important challenge as a medical test must apply to all potential patients or, if not, it must be very clear for whom the medical test is informative,” John Krystal said. • (Psychiatric News, Oct 2, 2018)
  34. Machine Learning Can Enhance Psychiatric Practice Inform Machine learning can inform hypothesis testing for future clinical trials Complement Machine learning and RCTs actually complement one another Augment Machine learning not meant to replace RCTs and clinical guidelines Helps Machine learning helps for patients not covered by the RCT groups and clinical guidelines
  35. Neural Networks and Deep Learning Fei Jiang et al. Stroke Vasc Neurol 2017;2:230-243
  36. • N. is a 19 year old single caucasian female, who is attending university and works part-time as a cashier for a pizza restaurant. She has a history of recurrent depressive episodes, which starts in the late fall/early winter months, and remits during the spring. She uses light therapy every morning for 30 minutes, as without it, she would be unable to function. Although her family doctor has recommended that she consider antidepressant medication treatment, N. has adamantly refused, stating that it is unnatural, as she tries to ingest healthy and natural foods. When she becomes depressed, she has symptoms of low moods, anhedonia, excessive fatigue, excessive sleep, increased appetite, carbohydrate cravings, and weight gain. • Her family history is significant for her mother suffering from bipolar disorder, and who dies in a car accident when N. was only 12 years old. It is highly suspected that the car accident was a suicide. N. has refused to see a therapist, stating she does not need psychological help, and that she can deal with her mother’s death on her own. • After the death of her mother, she and her brother were sent to live with their estranged father and step-mother, and their infant daughter, N.’s half-sister. She is the oldest of 3 siblings. Suffice it to say, N. did not get along well with her father and step-mother, and was jealous of all the attention her infant sister got from the family. When N. became depressed in the winter months, she locked herself in her room, and would sleep constantly, not contributing to the chores the family needed. When she was not sleeping, she would leave and spend days with her friends, without any check-ins with her parents. • Eventually,she ran away from home after her graduation from high school, and was not found until her father shamed her on Facebook that his daughter was missing and is now a runaway. Humiliated, N. contacted her father and assured him she was safe. • She eventually enrolled in university, and only has cursory contact with her father and siblings. Her first year in university was difficult, and she almost failed her classes due to sleeping too much when her depression would worsen like clockwork in the fall and winter. • When spring would come around, she would have increased energy and could work for 12 hours a day and go out with her friends at night with only minimal sleep. She spends her summer doing lots of outdoor activities, as she likes to soak in the sun, and dreads each day after June 21st, when the sun exposure would decrease little by little with each passing day. • Diagnosis: Major depressive disorder, recurrent, in partial remission, with seasonal pattern. Rule-out bipolar disorder, given the history of hypomanic symptoms, and family history of bipolar disorder. The patient had only a partial response to the current dose of light therapy. • Treatment: 1) Will increase light therapy duration to 60 minutes every morning, from a 10,000 lux light box. Will need to monitor for any signs of mania, as light therapy may switch someone with an underlying bipolar disorder from depression to mania. 2) If no response in 1 to 2 weeks, then consider augmenting light therapy with cognitive behavioral therapy (CBT). 3) Recommend other helpful strategies for depression, including exercise, meditation, yoga, and outdoor time to increase sunlight exposure. 4) Treatment with antidepressant medications is a relative contraindication, as bipolar disorder has not been ruled out. 5) Return to clinic in 2 weeks for follow-up.
  37. Extractive summarization process utilizing supervised (multiclass) classification based on the ontology
  38. Extractive summarization process utilizing supervised (multiclass) classification based on the ontology
  39. Carlo Carandang- Data Scientist and Psychiatrist Email: carlo@karvit.ca

Editor's Notes

  1. 1000 GB = 1 TB
  2. The road map from clinical data generation to natural language processing data enrichment, to machine learning data analysis, to clinical decision making. EMR, electronic medical record; EP, electrophysiological.
  3. Graphical illustration of unsupervised learning, supervised learning and semisupervised learning.
  4. Focus on Accuracy, Recall, and Precision. Also Sensitivity and Specificity- use example of algorithm that predicts suicide: want high Sensitivity, while optimizing that may reduce specificity, but that’s ok, as better to have a False Positive (unnecessary hospitalization) than a False Negative (discharge the patient from the ER and the patient suicides).
  5. With large volumes of health care data comes the research area of computational phenotyping, making use of techniques such as machine learning to describe illnesses and other clinical concepts from the data itself. The "traditional" approach of using supervised learning relies on a domain expert, and has two main limitations: requiring skilled humans to supply correct labels limits its scalability and accuracy, and relying on existing clinical descriptions limits the sorts of patterns that can be found. For instance, it may fail to acknowledge that a disease treated as a single condition may really have several subtypes with different phenotypes, as seems to be the case with asthma and heart disease. Some recent papers cite successes instead using unsupervised learning. This shows great potential for finding patterns in Electronic Health Records that would otherwise be hidden and that can lead to greater understanding of conditions and treatments. 
  6. Neural network One can think about neural network as an extension of linear regression to capture complex non-linear relationships between input variables and an outcome. In neural network, the associations between the outcome and the input variables are depicted through multiple hidden layer combinations of prespecified functionals. Deep learning: a new era of ML Deep learning is a modern extension of the classical neural network technique. One can view deep learning as a neural network with many layers Rapid development of modern computing enables deep learning to build up neural networks with a large number of layers, which is infeasible for classical neural networks. As such, deep learning can explore more complex non-linear patterns in the data. Another reason for the recent popularity of deep learning is due to the increase of the volume and complexity of data. Different from the classical neural network, deep learning uses more hidden layers so that the algorithms can handle complex data with various structures.27 In the medical applications, the commonly used deep learning algorithms include convolution neural network (CNN), recurrent neural network, deep belief network and deep neural network.
  7. An NLP pipeline comprises two main components: (1) text processing and (2) classification. Through text processing, the NLP identifies a series of disease-relevant keywords in the clinical notes based on the historical databases.42 Then a subset of the keywords are selected through examining their effects on the classification of the normal and abnormal cases. The validated keywords then enter and enrich the structured data to support clinical decision making. The NLP pipelines have been developed to assist clinical decision making on alerting treatment arrangements, monitoring adverse effects and so on. Furthermore, the NLP pipelines can help with disease diagnosis. 
Advertisement