DTU researchers hope to shorten waiting lists, so children can get answers and treatment sooner. And they hope their findings can help treat disorders like OCD directly.
Children and young people often wait months to get examined and treated by psychiatrists. But new technology in the form of AI might help things along.
At DTU Compute Associate Professor Line Clemmensen uses statistics and data analysis in combination with machine learning – a form of artificial intelligence – to help psychiatric clinicians in their work with children, a method called assistant AI.
Line Clemmensen has received several grants to aid her research, where she along with her colleagues work closely with researchers at the Children and Youth Psychiatric Center at Region Hovedstaden Psychiatry.
Her newest project, supported by the Lundbeck Foundation, looks at shortening waiting time to diagnosis and treatment of children and young people. In another project, supported by Novo Nordisk foundation, she is attempting to understand if AI can help in the treatment of children and young people with the mental disorder OCD, who suffer from obsessive thoughts and compulsive behaviour.
“The lack of resources in psychiatry is an enormous societal challenge. If we don’t have enough free hands available in psychiatry, it would be amazing if we could develop AI tools which could free up more hands. This could make diagnosis more efficient and help staff focus on treating and supporting children, young people, and their families,” says Line Clemmensen.
Reducing waiting times
To reduce the time children have to wait to get diagnosed and treated, Line Clemmensen and her colleagues are developing a machine learning model which can look through video recordings, made while doing health screening for psychiatric disorders.
"The lack of resources in psychiatry is an enormous societal challenge. If we don’t have enough free hands available in psychiatry, it would be amazing if we could develop AI tools which could free up more hands. This could make diagnosis more efficient and help staff focus on treating and supporting children, young people, and their families"
Associate Professor Line Clemmensen, DTU Compute
The model for this needs to be adapted to register the typical signs of the disorder in children so it can assist psychiatrists in evaluating which children need help, so the screening time is reduced.
“It’s not that simple because children are different to adults. They have no trouble listening and paying attention even though they don’t maintain eye contact with adults. Children also speak differently. They might be shy or mumble. Other times the quality of the video is poor or the camera has been placed in such a way that it is difficult for the machine learning models to register what is happening. But we believe we can solve these challenges,” Line Clemmensen says.
The AI-screening can also be used for quality control. Evaluating each other’s work is normal in psychiatry and the model could make this process easier as well, she explains:
“It is important to stress that we aren’t talking about letting machine learning do the work alone. It is a supplement to the health personnel to help them save time, which can be freed up to taking care of patients,” says Line Clemmensen.
Helping children with compulsive behaviour
Line Clemmensens machine learning research also focuses on how to help children suffering from the mental disorder OCD. These children are locked in a cycle of compulsive behaviour and obsessive thoughts where they keep repeating acts, often out of fear of something terrible happening.
In the pilot project funded by Novo Nordisk Foundation, the researchers test the effect of a special bracelet which could help children cope with OCD.
The bracelet sends data about their pulse, sweat and activity levels to a computer where the machine learning model, developed by the researchers, tries to predict when an OCD episode is about to happen. The computer then automatically sends a message to the parents that their child needs support to avoid the OCD-incident – or to the child itself to remind it how it can avoid doing the compulsive act.
Aimed at both children and parents
Normally children and young people receive therapy once a week and are supposed to work on changing their habits at home between therapy sessions, with help from their family.
“At first, we will study if we can detect OCD-incidents and next we will do targeted interventions. What is special about OCD is that the parents are often entangled in the OCD-cycle because they can see that the compulsive acts help the children in the moment. But the treatment is aimed at avoiding adaptation to the OCD by the family. They need to help children fight the OCD. A message to the parents or the young person can help them remember what was agreed upon in the therapy sessions,” explains Line Clemmensen.
A part of treating OCD also means agreeing to exposing the patients to things that provoke discomfort to show that even though it is uncomfortable, it isn’t dangerous. These tests are also performed by the family at home to give them a sense of taking back control of their lives.
“It’s one thing to do this in a clinical setting with a therapist. But they are also asked to do these exercises at home. And that is where our model can help them see how the reaction really was, which can sometimes be hard,” the DTU researcher says.
Not ready for reality yet
AI is heading into all parts of society, and it is natural for psychiatry to begin using the technology, explains senior researcher Nicole Nadine Lønfeldt at Region Hovedstaden’s Children’s psychiatric centre:
“We imagine the AI-driven diagnostic and psychotherapy tools we are developing will function as assistive devices for clinicians to make their jobs more efficient and objective, so that we can help more families.”
AI is already widely used in neuroscience research using MRI and machine learning is used to find new diagnostic categories and to predict concrete results like readmission. So far the use of artificial intelligence in mental health is experimental and have not been implemented practically in real world clinics.
“A great barrier in implementing AI-tools in psychiatric treatment is collecting the necessary evidence required to justify it. After developing the AI tools, we need randomised controlled experiments. We also need to be wary about what type of jobs we hand over to machines, because clinicians need to retain certain skills and to acquire knowledge without the help of AI. And there are also privacy concerns and risk of bias,” says Nicole Nadine Lønfeldt and continues:
“We also need to build trust in AI by developing tools with clinicians and patients where we have the opportunity to look at how the AI works, which is also important when training staff in using these tools.”
- To translate measurements of behaviour into the math used in machine learning models, Line Clemmensen and her colleagues start with already existing mathematical models. They are open source and already trained to examine the amount of time a person in a video is happy, angry, negative, positive, sad, has eye contact etc.
- The challenge is to choose the best models and to combine and adapt them to the task, so that the model looks at what is required to ascertain the mental condition of the patient. Nicoline Nadine Lønfeldt and her colleagues at Region Hovedstaden help the DTU team with recognising relevant clinical behaviour.
- The model needs to work even when the video quality is poor, noisy or if the camera is placed differently, to avoid wrongful reading of the patient. The models can also read biometrics data from the OCD-patients to register how the body reacts.