Machine Learning- the New Psychiatrist?
Written by Ju Wan Shin '26
Edited by Jacqueline Cho '24
25 key functional connections important for MDD diagnosis, color-coded according to the intrinsic network .
Can pathology extend to the mind? This is a debate that has long characterized the lengthy history of psychiatric conditions. The 21st-century consensus is that such conditions are rooted in disordered brain networks. However, linking specific alterations in the brain's functional connectivity to certain mental disorders has been an unsolvable challenge— until now. A biotech startup XNef Inc. will apply for official approval in Japan to use machine-learning software to support the diagnosis of major depressive disorder (MDD).
Current MDD diagnosis protocols are rather problematic. As diagnostic criteria used most commonly in clinical practice and research such as the Diagnostic and Statistical Manual of Mental Disorders or the International Classification of Disease rely heavily on the subjective assessment of symptoms by psychiatrists, there is very low concordance between individual psychiatrists in their methods of assessing MDD, compromising their validity. Thus, the existence of a supporting tool enabling objective diagnosis is crucial in providing more well-supported and accurate assessments.
“There is a critical and internationally recognized need to adopt a biological approach for mental-health diagnostics,” asserts psychiatrist Yuki Sakai, a leading XNef scientist and one of the company’s founders. Sakai also emphasizes that using a biologically based classification technique will help patients conceptualize their symptoms as a disease of the brain network, not a result of their own mental weakness .
A machine-learning software devised by XNef employs resting-state functional MRI imaging (rs-fMRI), a non-invasive and quick procedure that enables investigation of whole-brain functional connectivity, from which it then calculates the probability of MDD . The so-called 'XNef classifier' was not the pioneer of merging machine learning with fMRI data, however, as several precursors were constructed from images taken from a small number of participants at a single imaging site. The problem with the true pioneers of the mechanism? "A multisite dataset with multiple disorders raises difficult problems that are not present in a single site–based dataset", such as the "differences due to scanner type, imaging protocol, and patient demographics, even when a unified protocol was determined," explains Sakai .
These variations hinder the identification of generalizable brain markers for specific mental health conditions in a multi-site, multi-participant dataset. In response, XNef developed a harmonization method which smooths out the data by removing site-specific irregularities. Applying their harmonization technique, XNef identified markers for specific psychiatric conditions, such as MDD .
Another key application of brain network markers that is worthy of attention is the opportunity for drug development. In recent decades, progress in psychiatric disorder drug discovery has slowed down significantly; due to the high costs involved and the frequent unsuccessful outcomes of the randomized control studies, many pharmaceutical companies withdrew from the drug-discovery process. The main difficulty stems from the fact that mental illness is a syndrome and that patients cannot easily be placed into biologically uniform groups. Now, with the stratification of patients into biologically uniform subtypes with the assistance of brain network markers, clinical trials can focus on drugs that target specific subtypes to exponentially increase the rate of success.
Nature Research Custom Media. Supporting depression diagnosis through machine learning [Internet] [cited 2023 Apr 11]. Available from: https://www.nature.com/articles/d42473-022-00472-9.
Sakai Y, Yamada T, Yahata N, Kunimatsu A, Okada N, Itahashi T, et al.Generalizable brain network markers of major depressive disorder across multiple imaging sites. PLOS BIOLOGY [Internet]. 2020 [Cited 2023 Apr 11]. DOI: 10.1371/journal.pbio.3000966
Yamashita A, Yahata N, Itahashi T, Lisi G, Yamada T, Ichikawa N, et al. Harmonization of resting-state functional MRI data across multiple imaging sites via the separation of site differences into sampling bias and measurement bias. PLOS BIOLOGY [Internet]. 2019 [Cited 2023 Apr 11]. DOI: 10.1371/journal.pbio.3000042
Tanaka S, Yamashita A, Yahata N, Itahashi T, Lisi G, Yamada T, et al. A multi-site, multi-disorder resting-state magnetic resonance image database. Scientific Data [Internet]. 2020 [Cited 2023 Apr 11]; 8, 1- 16. DOI: s41597-021-01004-8