Head: Univ.-Prof. Dr. Nikolaos Koutsouleris
Teammembers: Dr. Dominic Dwyer, Riya Paul, Adyasha Khuntia, Rachele Sanfelici MSc, Maria Fernanda Urquijo MSc, Ömer Faruk Öztürk M.D., Dr. med. David Popovic, Santiago Tovar M.D., Adriana Herrera M.D., Anne Ruef, Mark Sen MSc, Susanne Miedl
The Neurodiagnostic Applications Section was founded in 2013 by Prof. Koutsouleris with the aim of developing machine learning methods for diagnostics and prediction in psychiatry. Since this time, the team has grown to 30 employees from different fields of medicine, psychology, and computer science. Promising research results from various projects (total funding > 2.5 million euros) have already been presented and published in more than 40 scientific articles. The team leads the EU-funded project PRONIA (Prognostic Tools for Early Psychosis Management) that has recruited a clinical cohort of more than 1500 test persons who were extensively clinically and biologically characterized. In addition, decisive progress has been made in establishing an efficient scientific infrastructure such as the development of a system for the management of clinical data and the production of machine learning software (NeuroMiner; www.pronia.eu/neurominer).
The Neurodiagnostic Applications Section is a leading member of several national and international research networks, such as the two large-scale NIH projects HARMONY (MH081902) and NAC-PS (PA-16-160), with the aim of using data integration to advance the development of machine learning methods for psychiatric prediction. In addition, junior lab leaders (Dr. Kambeitz & Kambeitz-Ilankovic) run several local research projects on personalized neuro-cognitive training, cannabis-induced psychoses, and the computational modelling of brain functional changes in psychoses.
NeuroMiner is a free machine learning software written by Prof. Nikolaos Koutsouleris in MATLAB, developed since 2009, that facilitates research into better tools for precision medicine. It provides a wealth of state-of-the-art supervised Machine Learning techniques, such as linear and non-linear support-vector machines, relevance vector machines, random forests, and gradient-boosting algorithms. It also comes with numerous dimensionality reduction methods and feature selection strategies that allow finding optimal combinations of predictive features for the user’s given prediction problem.
It can be operated with little coding experience as it is fully menu-driven which allows using NeuroMiner in server/remote environments with no or limited graphical interfaces. Furthermore, the NeuroMiner interface facilitates standardized parameter setup, storage, and dissemination across research labs, which is an essential requirement for more robust and generalizable predictive models.
NeuroMiner is constantly updated by the Section for Neurodiagnostic Applications (SNAP) in Psychiatry at the Department of Psychiatry and Psychiatry of Ludwig-Maximilian-University and can analyze all tabular data format that are stored in numeric MATLAB format, or alternatively in CSV or Microsoft Excel spreadsheets. It supports the in-depth analysis of 3D voxel-based neuroimaging data. These different data sources can be analyzed separately or be combined using a variety of data fusion approaches ranging from data concatenation, over bagging to more advanced stacking, or sequential data integration techniques.
In high-performance computing environments, researchers can exploit the SGE-/SLURM-based parallelization functionality of NeuroMiner to tremendously speed up the model training, cross-validation and visualization procedures.
The NeuroMiner Machine Learning School was first conducted in 2018. NeuroMiner is a machine learning software written by Prof. Nikolaos Koutsouleris in MATLAB that can be operated with little coding experience with a text-based menu system. This gives clinical researchers access to cutting-edge tools that can be used for predictive medicine and has been used for multiple publications.
Goal: The aim of this school is to introduce the participant to psychiatric machine learning with a specific focus on diagnostic and prognostic neuroimaging.
Structure: Fundamental machine learning concepts are introduced in lectures and the participant is guided through practical tutorials using software that does not require programming experience. By the end of the course, the participant will be able to plan and implement his own machine learning analyses with multiple types of data (e.g., clinical, biological, and neuroimaging).
Highlights of the school:
- Progression from data entry to advanced techniques (e.g., fusion and stacking)
- Hands-on tutorials using own computers in small groups guided by a tutor (n=10)
- Guest lecture from a machine learning expert
- A machine learning competition on the last day
- At the end of the machine learning school, the participant will be able to plan and conduct its own NeuroMiner analyses using varied data
Previous Machine Learning Schools experiences
- Machine Learning School October 2018
It was conducted in the LMU psychiatric clinic on Nußbaumstraße in Munich.
35 Participants were registered
The course had an excellent rating
Medical course participant: “I feel confident now”
- It was conducted by the ECNP Neuroimaging Network, the Section for Neurodiagnostic Applications, and the Max Planck Fellow Group for Precision Psychiatry.
- 124 Participants were registered for the general course, 50 participants attended specific tutorials conducted by 5 instructors.
- The course had an excellent rating
Next Machine Learning School:
On October 2021 is intended to be conducted the next Machine Learning School. Further information will be posted during the following months
Adriana Herrera M.D.
Machine Learning Engineer
Mark Sen MSc
PRONIA "Personalized Prognostic Tools for Early Psychosis Management" is a research project funded by the European Union to significantly improving the early detection of psychotic illness. The project aims to produce an easily accessible prognostic service for early illness detection. This forecasting system will use clinical information, cognitive data, neuroimaging, and DNA, RNA, and metabolomics.
PRONIA is comprised by a professional, motivated, and international team who develop computer models that can assess the risk of psychosis development, illness outcomes, and a range of other critical diagnostic and prognostic endpoints. In the future, these procedures will help doctors and psychologists to better assess and treat the course of the disease. For more information on this project, see PRONIA.
PNKT "Personalized neuro-cognitive training in depression and psychosis (high risk)" is a project in which researchers are working on the development of personalized neuro-cognitive therapies based on an individualized stratification of the risk for psychosis and a reduced functional level.
In addition to the typical symptoms of affective and non-affective psychosis, affected patients regularly suffer from cognitive impairments. Deficits in cognitive abilities in patients with psychosis are closely related to their quality of life and therefore represent a special focus in the rehabilitation of those affected. To date, it has been shown that computer-assisted neuro-cognitive training can be successfully used to reduce these cognitive deficits and at the same time improve the psychosocial function level of patients. However, previous studies show that only subpopulation of these patients offers a satisfactory response to cognitive training. One aim of the study is, therefore, to predict the positive response to this measure in order to offer cognitive training to those patients who will most likely benefit from it.
In the PNKT study, patients participate in a detailed clinical, neuro-cognitive, and imaging survey, followed by several weeks of computer-assisted cognitive training. Participation in the study ends with a follow-up examination. On the basis of multivariate pattern analyses, it is hoped that it will be possible to determine which data patterns are best for predicting whether and under what conditions a patient will benefit from cognitive training.
CIP - Cannabis-induced psychosis - is a DFG-funded study aiming to identifying individuals with cannabis-induced psychosis at an early stage and, in the context of a longitudinal imaging study, to find out in which affected individuals the psychotic symptoms remit spontaneously after abstinence and which patients are likely to develop a permanent form of psychosis.
Recent studies have shown that up to 50% of all patients with cannabis-induced psychosis develop a form of psychosis in the long term. Thus, patients with cannabis-induced psychosis represent the population with the highest known risk of permanent forms of psychosis. However, the neurobiological fundamentals of the connection between psychosis and cannabis use have hardly been understood so far. Nevertheless, this is urgently needed in order to identify the individuals with an increased risk of psychosis already at an early stage and to prevent the development of the disease and to positively influence the course of the disease.
In the CIP study, patients with cannabis-induced psychosis are tested for baseline and follow-up after 9 months by clinical and neuropsychological tests, as well as by using structural, functional and molecular imaging. The data is then evaluated by multivariate pattern analysis to identify brain patterns associated with clinical and neuropsychological variables.
TYPIA is a DFG-funded bicentric study conducted at the University of Bonn and the LMU Hospital in Munich. The aim of the study is to find out why not all people with highly pronounced schizotypic personality traits develop psychosis and the factors that protect against the development of psychosis. The project is based on observations of that suggest a continuum of psychosis symptoms between the individuals with no symptoms and populations of patients with psychosis.
Within the framework of TYPIA, test persons with high positive or negative schizotypic characteristics, control subjects with low schizotypic characteristics, and patients suffering from psychosis for the first time in the last three years are examined. The age range for all groups is between 18 and 40 years. The participation includes examinations of personality, cognition, DNA, a clinical interview, as well as images of brain structure and function using magnetic resonance imaging (MRI).
PhD position in transdiagnostic biomarker investigations (03.12.2020):
We invite masters degree to apply to this unique opportunity of working at the intersection of psychotic and affective disorders with focus on translational machine learning. For further information please refer to the this MPI link.
Post-doc position in interpretable machine learning (03.12.2020):
We invite finished PhD students to apply to this unique opportunity of working at the intersection of psychotic and affective disorders with focus on implementation of cutting-edge machine learning solutions. For further information please refer to the this MPI link.
For the second time in September 2020, the NeuroMiner Machine Learning School was conducted by the ECNP Neuroimaging Network, the Section for Neurodiagnostic Applications, and the Mx Planck Fellow Group for Precision Psychiatry. A total of 124 participants were registered in the 4 day course and 50 participants attended specific tutorials conducted by 5 instructors. It was a successful event with excellent feedback from the participants
- Koutsouleris, N., Dwyer, D. B., Degenhardt, F., Maj, C., Urquijo-Castro, M. F., Sanfelici, R., . . . Meisenzahl, E. (2020). Multimodal Machine Learning Workflows for Prediction of Psychosis in Patients With Clinical High-Risk Syndromes and Recent-Onset Depression. JAMA Psychiatry. doi:10.1001/jamapsychiatry.2020.3604
- Dwyer, D. B., Kalman, J. L., Budde, M., Kambeitz, J., Ruef, A., Antonucci, L. A., . . . Koutsouleris, N. (2020). An Investigation of Psychosis Subgroups With Prognostic Validation and Exploration of Genetic Underpinnings: The PsyCourse Study. JAMA Psychiatry, 77(5), 523-533. doi:10.1001/jamapsychiatry.2019.4910
- Popovic, D., Ruef, A., Dwyer, D. B., Antonucci, L. A., Eder, J., Sanfelici, R., . . . Koutsouleris, N. (2020). Traces of Trauma: A Multivariate Pattern Analysis of Childhood Trauma, Brain Structure, and Clinical Phenotypes. Biol Psychiatry, 88(11), 829-842. doi:10.1016/j.biopsych.2020.05.020
- Sanfelici, R., Dwyer, D. B., Antonucci, L. A., & Koutsouleris, N. (2020). Individualized Diagnostic and Prognostic Models for Patients With Psychosis Risk Syndromes: A Meta-analytic View on the State of the Art. Biol Psychiatry, 88(4), 349-360. doi:10.1016/j.biopsych.2020.02.009
- Dwyer, D. B., Falkai, P., & Koutsouleris, N. (2018). Machine Learning Approaches for Clinical Psychology and Psychiatry. Annu Rev Clin Psychol, 14, 91-118. doi:10.1146/annurev-clinpsy-032816-045037
- Koutsouleris, N., Kambeitz-Ilankovic, L., Ruhrmann, S., Rosen, M., Ruef, A., Dwyer, D. B., . . . Borgwardt, S. (2018). Prediction Models of Functional Outcomes for Individuals in the Clinical High-Risk State for Psychosis or With Recent-Onset Depression: A Multimodal, Multisite Machine Learning Analysis. JAMA Psychiatry, 75(11), 1156-1172. doi:10.1001/jamapsychiatry.2018.2165
- Koutsouleris N, Kahn RS, Chekroud AM, Leucht S, Falkai P, Wobrock T, Derks EM, Fleischhacker WW, Hasan A. (2016). Multisite prediction of 4-week and 52-week treatment outcomes in patients with first-episode psychosis: a machine learning approach. Lancet Psychiatry, 3(10):935-946. doi: 10.1016/S2215-0366(16)30171-7.
- Koutsouleris N, Meisenzahl EM, Borgwardt S, Riecher-Rössler A, Frodl T, Kambeitz J, Köhler Y, Falkai P, Möller HJ, Reiser M, Davatzikos C. (2015). Individualized differential diagnosis of schizophrenia and mood disorders using neuroanatomical biomarkers. Brain. 138(Pt 7):2059-73. doi: 10.1093/brain/awv111.
- Koutsouleris N, Meisenzahl EM, Davatzikos C, Bottlender R, Frodl T, Scheuerecker J, Schmitt G, Zetzsche T, Decker P, Reiser M, Möller HJ, Gaser C. (2009). Use of neuroanatomical pattern classification to identify subjects in at-risk mental states of psychosis and predict disease transition. Arch Gen Psychiatry. 66(7):700-12. doi: 10.1001/archgenpsychiatry.2009.62.
As the leader of the PRONIA consortium, the neurodiagnostics laboratory closely collaborates with the following groups:
- The University of Basel, Department of Psychiatry, Prof. Stefan Borgwardt and Prof. Anita Riecher-Rössler
- The University of Cologne, Department of Psychiatry, Prof. Joachim Klosterkötter
- The University of Birmingham, Department of Psychology, Prof. Stephen Wood
- The University of Turku, Department of Psychiatry, Prof. Raimo Salokangas
- The University of Udine, Department of Psychiatry, Prof. Paolo Brambilla
- The University of Melbourne, Department of Psychiatry, Prof. Christos Pantelis
- General Electric Deutschland & GE Healthcare
As part of the HARMONY consortium, the neurodiagnostics laboratory closely collaborates with the following groups:
- Yale University, Department of Psychology, Prof. Tyrone Cannon
- King's College London, Institute of Psychiatry, Prof. Philip McGuire
- The University of Pennsylvania, Department of Psychiatry, Prof. Raquel Gur
An ongoing collaboration is also with:
- The University of Pennsylvania, Center for Biomedical Image Computing and Analytics, Prof. Christos Davatzikos