Skip to main content

Developing Machine Learning Classifier Models for Eye Movements to Diagnose Major Psychiatric Disorders

Periodic Reporting for period 1 - SACCSCAN-IA-ML (Developing Machine Learning Classifier Models for Eye Movements to Diagnose Major Psychiatric Disorders)

Reporting period: 2017-09-29 to 2018-09-28

Mental ill health is now recognised as the largest cause of short and long term disability worldwide costing the global economy US$2.5T (€798 billion in Europe), with a projected increase to over US$6T by 2030. Approximately 1 in 5 of us (more than 450 million people) experience mental health problems during our lifetime with about 5% suffering from a serious mental illness such as schizophrenia, bipolar, and major depressive disorders. Illness specific treatments are available that allow patients to resume normal functioning in society but physicians are struggling to make accurate diagnosis, match therapy to condition, and provide timely care, affecting more than 50% of patients who don’t receive adequate care.

The current gold standard is based on interviewing patients over several months or years to document their behaviour and symptoms with misdiagnosis occurring 50% of the time. When patients’ history, symptoms and behaviour don’t meet the criteria set out in the diagnostic manual, it may take up to 10 years to diagnose the illness. Symptom based clinical evaluation requires significant amount of time, expertise, hence cost. There also remains the uncertainty as to whether an illness can in any event be diagnosed without the full diagnostic criteria being met. Delays in receiving a diagnosis can significantly impede delivery of the most effective treatment plan, exposing the patient to risk of further deterioration in well-being, reduction in quality of life leading to job loss, family breakdown, and self-harming.

The development of simple objective tests for the major psychiatric disorders that are stable, sensitive and specific enough to be of clinico-diagnostic and treatment value would be of enormous benefit. Decades of effort have been spent attempting without success to identify assays that could be used as diagnostic markers in psychiatry. Attempts to find radiological and serological assays have proved especially frustrating. Although abnormalities of eye movements were first observed in unmedicated psychotic patients over 100 years ago, they have never been considered sensitive or specific enough to be of diagnostic value at the level of the individual patient, and as a result, the observations have been generally ignored and neglected. Modern eye tracking devices and sophisticated analytical techniques have reinvigorated interest in the potential of eye movements to assist psychiatrists and healthcare professionals. We are the first to discover that different eye-movement tests give substantially more information about underlying neural pathways and realised that a combination of performance measures and tests would be key to developing a psychiatric biomarker with the required sensitivity and specificity. This has led to the development of the patent protected SaccScan eye test which has been demonstrated to detect schizophrenia with better than 95% accuracy and is being extended with the same precision to bipolar disorder and major depression illnesses. The test can be performed within 30 minutes with simple desktop equipment and results produced over the internet.
A working machine learning model for schizophrenia was used as a starting point in this project and was extended to bipolar disorder and major depressive disorder. The SaccScan software diagnostic tool will scale up systematically adding more illness categories and diagnostic insight to its range of features as the eye movement clinical reference database of patients and controls increases over the years. We developed an accurate and reliable estimation method of the entire combined system to differentially diagnose schizophrenia, bipolar disorder and major depressive disorder. We have also prototyped an effective strategy for visualizing results directly applicable to new patient data. Tools were introduced within the organisation to design, develop, and monitor the system as a medical device in compliance with regulatory requirements for CE marking. Classification of patients blind to diagnosis into sub-groups due to their self-similar eye movement measures is a validated and powerful technique to discover phenotypic clusters along a trait continuum. The outcome of these tasks will extend the SaccScan test to include other major psychiatric disorders such as bipolar disorder and major depression. Such an accomplishment followed by regulatory approval will make the SaccScan proposition highly attractive to clinical end users both in primary care services, community health centres as well as hospitals worldwide.
We have demonstrated at proof-of-concept that sophisticated machine learning (ML) methods are required to model eye movement phenotypes. This will be a significant innovation, entirely novel in the field of eye movements for diagnosing psychiatric disorders. Our principal aim is to develop the most robust classifier models on our expanding clinical reference database for a range of mental illnesses starting with the three most common major psychiatric disorders: schizophrenia, bipolar disorder and major depressive disorder. Although practitioners will be the immediate end users of the test, health care systems will also benefit directly from the test from efficiencies in patient treatment pathways, and the need is also being driven by requests from the patient community. Our health economics assessment of SaccScan showed savings of €40k per patient with suspected schizophrenia alone. These savings are from early detection leading to fewer patients receiving suboptimal treatment and requiring hospitalisation. Results from the model support findings from the Centre for Economics of Mental Health in 2010, which reported €10 return (of which €2.6 would accrue to the health care system) on every €1 invested in early detection of major psychiatric disorders, especially psychosis (schizophrenia and related disorders), realised from reduction in outpatient appointments, inpatient admissions and forensic costs, and similar findings were also published by research groups in Italy and Denmark .