e-Mindfulness, Psychotherapy and Technology Lab �(e-MPAT LAB) – the scope of possible projects

Warsaw, Poland

Paweł Holas¹ Karol Chlasta² Szymon Szumiał¹

​

​

​

​

​

1. University of Warsaw, Faculty of Psychology
2. Polish-Japanese Academy of Information Technology

PLAN

• Problem (Background):
• High prevalence of psychiatric disorders
• Diagnosis in mental health (Human vs Computer)
• Lack of validated methods in psychotherapy effectiveness research

​

• e-MPAT Lab projects and needs

​

• Example of Study: Detection of Dementia in Speech
• Future directions

​

​

BACKGROUND I – Emotional disorders:�Anxiety and depression

• Depression is the leading cause of disability worldwide. *
• Almost 75% of people with mental disorders remain untreated in developing countries
• Almost 1 million people taking their lives each year.
• The WHO reports that anxiety disorders (AD) are the most common mental disorders worldwide with specific phobia, and social phobia being the most common anxiety disorders.
• In the U.S., AD affect 18.1% of the population every year.
• AD are highly treatable, yet only 36.9% of those suffering receive treatment.

�

* WHO, * * National Institute of Mental Health,

BACKGROUND I - DEPRESSION

• The lifetime prevalences of major depressive disorder is 20.6%, respectively, with most being moderate (6-7 symptoms) or severe (8-9 symptoms) and associated with comorbidity and impairment (in US).
• Around 17.3 million adults in the U.S. had experienced at least one major depressive episode in the last year (6.7% of adults in the U.S
• Depressive symptoms are present in approximately 15% of older adults
• There are several geriatric-specific variants of depression with “depression-executive dysfunction syndrome” the most common and easily confused with dementia

* National Institute of Mental Health

BACKGROUND I - Challenges of Population Ageing

• The aged population is currently at its highest level in human history
• The number of people aged 60 years and over has tripled since 1950, reaching to 16% in 2050
• Leading to new medical challenges (WHO 2011)

​

• Dementia is the second most common mental health problem in aged population, with about 5% prevalence worldwide, but it is the first leading cause of disease burden.

BACKGROUND II - DEMENTIA

• By the age of 85 years, between 25 and 50 percent of people display symptoms of dementia

​

• Within the US, approximately 5.7 million people are living with dementia of the Alzheimer’s type (AD)

​

• Alzheimer's disease accounts for 60%–70% of these cases, followed by vascular dementias

19-1-27

BACKGROUND III - PSYCHOTHERAPY

19-1-27

Psychotherapy – main questions

• What therapy, and by whom, is most effective for a particular person with this particular problem, as well as under what circumstances and how it occurs.

​

Paul, 1969

e-MPAT LAB – directions we take:�Development in psychotherapy!?

• What most explains the effect variance => therapist's variance
• Developing methods and tools to assess and monitor therapy effects, especially measures of the above key phenomena (covenant, engagement...)
• Developing and evaluating monitoring and feedback tools based on new technologies (NLP, Neural Networks, ML) => e-MPAT LAB
• Developing Feedback-Informed Therapy

Psychotherapy Research �e-MPAT Lab:

• To study process and therapist-dependent factors that are responsible for effectiveness of therapy and for increase of engagement (lower attrition).
• Analysis of language markers based on NLP and machine learning (ML/AI)
• Analysis of non-verbal audio signals (physical speech properties) and visual signals (e.g. face) – with ML/AI
• In addition to outcomes therapy measurement and psychophysiological (HRV) assessment

​

e-MPAT lab projects:

• Development of different models (traditional machine learning and deep machine learning) of classification and regression to support the diagnosis of mental disorders, and to evaluate the therapy process.
• Evaluation of the effectiveness and mechanisms of text-based psychotherapy (i.e., instant messaging/texting intervention)
• Help in development of an application (web/mobile) collecting data for training machine learning models (speech in the field of speech content analysis - NLP, as well as physical properties of speech as sound signal, video) and implementing diagnostic tools.

​

​

Screening – diagnosis assistance with audio and visual markers

• Screening – Detection of deviations and trends in indicators of mental functioning
• Ability to detect and monitor disorders early and during treatment
• Analogy with online ECG in cardiology

​

Artificial Intelligence (AI), machine learning (ML): language (speech), sound, digital data

e-MPAT lab projects (cont.):

• Project in the field of Internet addictions: � Development and implementations of cognitive online task in gaming disorder (PsychPy), � Help in data collection and analyses of continous measurement of psychophysiological data (e.g. HRV) and ecological momentary assessment (EMA) data.

​

• A project of building a functional model of early dementia based on the Semantic Pointer Architecture Unified Network (SPAUN) model using the Python language.

e-MPAT Lab - our basic needs (1):�

1. We need software for transcription of audio/video recordings working in Polish language, preferably using Google engine
2. We need help programming the ability to search and count text categories in R Statistics and/or Python.
3. We need involvement in computations for extracting text categories from therapist and patient statements based on unsupervised machine learning methods
4. We need involvement in supervised learning-based computations linking extracted text categories to objectified measures of patient functioning, such as symptom severity.

​

​

e-MPAT Lab - our basic needs (2):�

5. We need help programming and creating a cognitive task (probably best in PsychoPy)
6. We need help developing an application (web/mobile) collecting data for training machine learning models
7. We need help in building a functional model of early dementia based on the Semantic Pointer Architecture Unified Network (SPAUN) model using the Python language.

​

​

CLINICAL CHALLENGES – study 1

Dementia - clinical diagnosis is mainly based on mental status examination and direct interview.

Sometimes it is very difficult to make a firm diagnosis of depression in the context of dementia, especially when the dementia is very advanced.

The clinician will occasionally choose to instigate treatment regardless of diagnostic certainty, weighing the possible benefits versus potential adverse outcomes of treatment.

CLINICAL DIFFERENTIAL DIAGNOSIS

​

Example of rsearch STUDY 1: AIM

• This study examined the automated de detection in speech using
• (1) VGGish model and Scikit-learn classifiers
• (2) Custom convolutional neural network
• We expected that classification of voice samples using (1) (2) architectures could give promising results in dementia detection

STUDY 1: METHODS

Participants from DementiaBank, ADReSS dataset (n = 78)

STUDY 1: METHODS

Two-stage architecture: VGGish model and Scikitlearn classifiers

Architecture diagram of custom PyTorch convolutional neural network (CNN)

STUDY 1: METHODS

VGGish

and Scikitlearn

• VGGish Feature Extractor Trained on YouTube Data
• PCA was used for dimensionality reduction, with PCA set to 128
• SVM, LinearSVM, Perceptron, MLP, 1NN

Custom PyTorch CNN

• 6 Conv1D layers using ReLU activation function
• Final (7th) output layer being a dense layer
• Two step training of our CNN model using a cross entropy loss function

STUDY 1: MAIN FINDING

• The proposed method produced a promising AD classification accuracy of
• 63.6% for a custom CNN with 7 layers
• 59.1% for a pre-trained VGGish model and scikit-learn classifiers.
• These results can be improved by retraining on more focused, larger datasets

Confusion matrix for the best performing CNN model that achieved accuracy of 63.6% on ADReSS data set.

STUDY 1: CONCLUSIONS

• Promising new direction for preliminary screening of subjects with dementia of the Alzheimer’s type by using short samples of their voice

​

• Both methods offer progress towards new and more effective speech technologies that could be used in an innovative screening system for dementia

FUTURE DIRECTIONS

• Modern investigative tools, such as neuroimaging etc. are very expensive
• A speech based (audio and scripts analyses) screening system can be used independently, or as an element of a more complex, hybrid, or multimodal solution
• Its use in effectiveness of psychotherapy and other types of treatment research is the promising direction for future…

Selected publications from the area of e-Mental Health

• Holas, P. (2021). I-mindfulness-based cognitive therapy (i-MBCT) in the treatment of COVID-19 related adjustment disorder. a RCT study with active control group. European Psychiatry, 64(S1), S100-S100.
• Chlasta, K., Holas, P., & Wolk, K. (2021). Computer-based detection of depression and dementia in spontaneous speech. European Psychiatry, 64(S1), S349-S349.
• Wołk, A., Chlasta, K., Holas, P. (2021). Hybrid approach to detecting symptoms of depression in social media entries. PACIS 2021 Proceedings. 192. https://aisel.aisnet.org/pacis2021/192.
• Chlasta, K., Wołk, K., & Krejtz, I. (2019). Automated speech-based screening of depression using deep convolutional neural networks. Procedia Computer Science, 164, 618-628.
• Chlasta, K., & Wołk, K. (2020). Towards computer-based automated screening of dementia through spontaneous speech. Frontiers in Psychology, 11, 4091.

​

​

​

THE END - QUESTIONS?�E-MPAT Lab members:

1. Pawel Holas,
2. Rafał Styła
3. Hubert Suszek
4. Szymon Szumiał
5. Patryk Roczon (MA, PhD candidate)
6. Antoni Korczak (PhD candidate)
7. Karol Chlasta (affiliate member)

email:

• pawel.holas@psych.uw.edu.pl

​