A recent study published in the renowned journal Nature Scientific Reports has shed light on the potential of the clock drawing test (CDT) as a reliable and affordable tool for diagnosing dementia. The groundbreaking research, led by Ph.D. candidate Sabyasachi Bandyopadhyay from the intelligent Health Lab (i-Heal), uses a deep learning model called the relevance factor variational autoencoder (RF-VAE), to analyze digitized clock drawings to identify unique features associated with dementia.
The clock drawing test has long been utilized as a cost-effective screening method for cognitive impairments, including dementia. However, Bandyopadhyay’s study delves deeper into this test by employing advanced deep generative neural networks. By using the RF-VAE model, the research team was able to extract valuable insights from clock drawings obtained from various institutions.
The findings of the study highlight the significance of specific constructional features in clock drawings that had not been extensively studied in prior research. Notably, they identified that characteristics associated with dementia include small-sized clocks with a non-circular shape resembling an avocado, along with irregularly placed hands. These distinctive features proved highly informative in accurately distinguishing dementia patients from non-dementia individuals, achieving an impressive area under the receiver operating characteristic (AUC) score of 0.96 when combined with demographic information.
This pioneering study represents a significant advancement in the field of dementia diagnosis. The RF-VAE network’s remarkable accuracy in classifying dementia cases showcases the power of clock drawing as a simple yet effective tool for detecting cognitive impairments, particularly in individuals suspected of having mild cognitive impairment or dementia.
While previous attempts at classifying dementia using two-dimensional latent space VAE networks showed promise, this study built upon those findings by identifying a comprehensive set of independent and informative graphomotor features specific to clock drawing. These features exhibited outstanding performance in distinguishing dementia cases from controls.
Moving forward, the research team plans to expand the investigation to encompass post-surgical cognitive dysfunction, Parkinson’s disease, and other specific types of dementia. Furthermore, they aim to validate their findings by utilizing independent publicly available datasets. The ultimate goal is to develop explainable and effective diagnostic models for cognitive impairments through the pioneering work in generative feature learning using deep neural networks on clock drawing data.
- Sabyasachi Bandyopadhyay, Ph.D. candidate, J. Crayton Pruitt Family Department of Biomedical Engineering, University of Florida
- Parisa Rashidi, Ph.D., Associate Professor, J. Crayton Pruitt Family Department of Biomedical Engineering, University of Florida
- Jack Wittmayer, Department of Computer and Information Science and Engineering, University of Florida
- Patrick Tighe, MD, Associate Professor of Anesthesiology, Department of Anesthesiology, College of Medicine, University of Florida
- Catherine Price, Ph.D., Associate Professor, Department of Clinical and Health Psychology, College of Public Health and Health Professions, University of Florida