Gautham Sundar
Data is something that the world cannot do without in its current state. Every field that one can think of, whether it be a big multinational corporation or sports teams or even your local supermarket, all of them make use of data in some way or another. A useful way of analyzing all of this data is via machine learning (ML). The usage of data has also grown amongst mental health researchers with the increased use of social media, neuroimaging and smartphones (Chen et al., 2014) and as a result, the usage of machine learning is something that has become more prominent. In this post, we will go over what machine learning is, how it is being used in the field of clinical psychology and whether or not it is something that can be widely used in this field in its current form.
ML utilizes complex “statistical and probabilistic techniques to construct systems that automatically learn from data” (Shatte et al., 2019). This makes it easier to identify various patterns in data and as a result one can make more accurate predictions from various data sets. ML also consists of various types, some of which are supervised/unsupervised learning, active learning and deep learning. ML applications in mental health are various, but the focus here is on how ML helps with the detection and diagnosis of disorders.
Firstly, ML helps with the detection of disorders in a variety of ways. One way involves the use of personal sensing using sensors on everyday devices like smartphones, smartwatches and even analyzing data from an individual’s patterns on social media. Mental health professionals can get useful data on things like skin conductance, entropy, circadian rhythm and even the nature of posts on sites like Instagram and Twitter. They can then use this data as potential predictors for disorders like depression, schizophrenia and ADHD (Mohr et al., 2017). Another application of ML for detection involves the analysis of neuroimaging data. Studies have been done to assess people’s vulnerability to depression by analyzing data from functional magnetic resonance imaging (fMRI) scans, with an accuracy of 78.3% (Sato et al., 2015) and even make it easier for professionals to detect cases of psychosis (Koutsouleris et al., 2012). Such cases have shown that ML can be used to improve the detection of certain disorders and as a result, diagnose individuals as soon as possible or take measures to prevent the onset of these disorders.
Secondly, ML has also helped with making the process of diagnosis itself easier. Many ML models have been developed using neuroimaging data from fMRIs, electroencephalograms (EEG) and positron emission topography (PET) scans. This has led to ML being able to diagnose patients of schizophrenia and Alzheimer’s disease with an accuracy of 62% to 70% (Shatte et al., 2019). It was also shown to be able to accurately distinguish between individuals with similar symptoms, like those with autism spectrum disorders or epilepsy with an accuracy of greater than 95% (Bosl et al., 2017).
While ML can be extremely useful for diagnosis, as highlighted by the previous examples, it does come with a few challenges. Firstly, a challenge that all users of ML face is the fact that the data used to develop a ML model still needs to be very robust. ML is not a substitute for many of the conventional research methods that psychologists use, rather it can be used as a supplement that makes use of this data for various purposes. The use of ML also raises ethical questions surrounding privacy and the collection of sensitive data related to mental health. It is not clear whether a lot of the users of social media or devices like smartphones and smartwatches have given their consent for their activity or data to be collected and analyzed in this manner. It is also not clear if the ML models being developed for such studies have strong privacy measures in place so that data is not easily misused.
While these are challenges for the use of ML at large, there are also certain limitations when it comes to ML and its use for diagnosis. Firstly, all of these diagnoses were made in a research context and one cannot be sure if this would work for the public as the data being collected will inevitably be different from the ones used in the lab. Secondly, there seems to be no standardized ML model that has been developed to analyze data collected by professionals. This seems to be worrying as a lot of the data that psychiatrists collect before making a diagnosis is through standardized questionnaires or interviews that allows them to see if an individual meets the criteria for a particular disorder. Most of the research done in ML for mental health uses different models and there does not seem to be one model that seems to work across different datasets and is a potential avenue for further research. Finally, a lot of the research that has been done is disproportionate towards the analysis of imaging data and it is not clear whether ML is as effective in other data types like speech patterns and sensor data. This means that ML in its current state is of most use to those who work with imaging data, and many researchers might not have access to technology like fMRIs, EEGs or PET scanners.
To conclude, ML has been seen to be very useful for the purposes of detecting and taking preventive action against disorders before their onset and also for accurately diagnosing individuals who are suffering from a disorder. However, ML does face a lot of challenges and as a result, it is still not something that can be used at a very large scale. Nevertheless, a lot of research being done using ML is very recent and many developments in its use are sure to follow.
References
Bosl, W. J., Loddenkemper, T., & Nelson, C. A. (2017). Nonlinear EEG biomarker profiles for autism and absence epilepsy. Neuropsychiatric Electrophysiology, 3(1). https://doi.org/10.1186/s40810-017-0023-x
Chen, M., Mao, S., & Liu, Y. (2014). Big Data: A Survey. Mobile Networks and Applications, 19(2), 171–209. https://doi.org/10.1007/s11036-013-0489-0
Koutsouleris, N., Borgwardt, S., Meisenzahl, E. M., Bottlender, R., Möller, H.-J., & Riecher-Rössler, A. (2011). Disease Prediction in the At-Risk Mental State for Psychosis Using Neuroanatomical Biomarkers: Results From the FePsy Study. Schizophrenia Bulletin, 38(6), 1234–1246. https://doi.org/10.1093/schbul/sbr145
Mohr, D. C., Zhang, M., & Schueller, S. M. (2017). Personal Sensing: Understanding Mental Health Using Ubiquitous Sensors and Machine Learning. Annual Review of Clinical Psychology, 13(1), 23–47. https://doi.org/10.1146/annurev-clinpsy-032816-044949
Sato, J. R., Moll, J., Green, S., Deakin, J. F. W., Thomaz, C. E., & Zahn, R. (2015). Machine learning algorithm accurately detects fMRI signature of vulnerability to major depression. Psychiatry Research: Neuroimaging, 233(2), 289–291. https://doi.org/10.1016/j.pscychresns.2015.07.001
Shatte, A. B. R., Hutchinson, D. M., & Teague, S. J. (2019). Machine learning in mental health: a scoping review of methods and applications. Psychological Medicine, 49(09), 1426–1448. https://doi.org/10.1017/s0033291719000151
Su, C., Xu, Z., Pathak, J., & Wang, F. (2020). Deep learning in mental health outcome research: a scoping review. Translational Psychiatry, 10(1). https://doi.org/10.1038/s41398-020-0780-3
Comments
Post a Comment