Ontology driven decision support for the diagnosis of mild cognitive impairment
This paper focuses on the development of an ontology for mild cognitive impairment (MCI). Alzheimer’s Disease (AD) can be diagnosed with reasonable accuracy at the stage of dementia, which is a point at which little can be done to achieve a favorable outcome. Hence there is a keen interest in being able to diagnose AD before the dementia stage is reached. Studies at Mayo’s Alzheimer’s Disease Research Center (ADRC) show that 8 out of 10 patients with MCI will convert to AD. Being able to accurately diagnose MCI can help with diagnosing AD pre-dementia stage. Prior methods for diagnosing MCI used observation based criteria which are subject to bias and could result in misdiagnosis. To remove the possibility of subjectivity resulting in misdiagnosis, an ontology for MCI containing specialized MRI knowledge about the cortical thickness of the brain structure was created to be used by clinical decision support systems when analyzing brain MRI scans for a patient.
The components of their framework consist of a MCI knowledge repository, an inference mechanism (rule sets extracted using machine learning algorithms), a feature obtaining process (measurements of the cortical thickness) and data processing mechanism . The inference mechanism uses the C4.5 algorithm and it was trained using MRI data for 187 MCI patients and 177 non-MCI patients who served as normal controls.
They obtained a sensitivity score of 80.2% and with 10-fold cross validation, they were able to show that it performed better than other algorithms like support vector machines (SVM) and Bayesian network (BN) and back propagation (BP) .
This is an extensive study using machine learning algorithms with MRI data for patient diagnosis. Validation using independent data sets would be important for clinical translation.
Write something here
- Zhang, X., Hu, B., Ma, X., Moore, P., & Chen, J. (2014). Ontology driven decision support for the diagnosis of mild cognitive impairment. Computer Methods and Programs in Biomedicine, 113(3), 781–791. http://doi.org/10.1016/j.cmpb.2013.12.023