Andreas Müller1, Sarah Vetsch1, Ilia Pershin1, Gian Candrian1, Gian-Marco Baschera1, Juri D Kropotov2, Johannes Kasper3, Hossam Abdel Rehim4, Dominique Eich5. 1. Brain and Trauma Foundation Grisons/Switzerland, Chur, Switzerland. 2. N.P. Bechtereva Institute of the Human Brain, Russian Academy of Sciences, St. Petersburg, Russia. 3. Praxisgemeinschaft für Psychiatrie und Psychotherapie, Lucerne, Switzerland. 4. Psychiatrie und Psychotherapie Rapperswil, Rapperswil, Switzerland. 5. Department of Psychiatry, Psychotherapy and Psychosomatics, University of Zurich, Zurich, Switzerland.
Abstract
Objectives: The electrophysiological characteristics of attention-deficit/hyperactivity disorder (ADHD) and recent machine-learning methods promise easy-to-use approaches that can complement existing diagnostic tools when sufficiently large samples are used. Neuroalgorithms are models of multidimensional brain networks by means of which ADHD patient data can be separated from healthy control data. Methods: Spontaneous electroencephalographic and event-related potential (ERP) data were collected three times over the course of 2 years from a multicentre sample of adults comprising 181 patients with ADHD and 147 healthy controls. Spectral power and ERP amplitude and latency measures were used as input data for a semi-automatic machine-learning framework. Results: ADHD patients and healthy controls could be classified with a sensitivity ranging from 75% to 83% and specificity values of 71% to 77%. In the analysis of the repeated measurements, sensitivity values of the selected logistic regression model remained high (72% and 76%), while specificity values slightly decreased over time (64% and 67%).Conclusions: Implementation of the system in clinical practice requires facilities to track affected networks, as well as expertise in neuropathophysiology. Therefore, the use of neuroalgorithms can enhance the diagnostic process by making it less subjective and more reliable and linking it to the underlying pathology.
Objectives: The electrophysiological characteristics of attention-deficit/hyperactivity disorder (ADHD) and recent machine-learning methods promise easy-to-use approaches that can complement existing diagnostic tools when sufficiently large samples are used. Neuroalgorithms are models of multidimensional brain networks by means of which ADHDpatient data can be separated from healthy control data. Methods: Spontaneous electroencephalographic and event-related potential (ERP) data were collected three times over the course of 2 years from a multicentre sample of adults comprising 181 patients with ADHD and 147 healthy controls. Spectral power and ERP amplitude and latency measures were used as input data for a semi-automatic machine-learning framework. Results:ADHDpatients and healthy controls could be classified with a sensitivity ranging from 75% to 83% and specificity values of 71% to 77%. In the analysis of the repeated measurements, sensitivity values of the selected logistic regression model remained high (72% and 76%), while specificity values slightly decreased over time (64% and 67%).Conclusions: Implementation of the system in clinical practice requires facilities to track affected networks, as well as expertise in neuropathophysiology. Therefore, the use of neuroalgorithms can enhance the diagnostic process by making it less subjective and more reliable and linking it to the underlying pathology.
Authors: Bettina L Serrallach; Christine Groß; Markus Christiner; Simon Wildermuth; Peter Schneider Journal: Front Neurosci Date: 2022-05-06 Impact factor: 5.152