Data-Driven Differential Diagnosis of Dementia Using Multiclass Disease State Index Classifier

Area of Science:

• Neuroscience
• Medical Informatics

Background:

• Clinical decision support systems (CDSSs) show promise for diagnosing neurodegenerative diseases.
• Accurate differential diagnosis of dementias is crucial for effective patient management.

Purpose of the Study:

• To evaluate the performance of the novel PredictND CDSS tool in differentiating between controls and various dementia types.
• To assess the accuracy of the multiclass Disease State Index classifier within the PredictND tool.

Main Methods:

• Utilized a subset of the Amsterdam Dementia cohort (504 patients).
• Integrated data from neuropsychological tests, cerebrospinal fluid samples, and MRI quantifications.
• Employed the multiclass Disease State Index classifier for differentiating five classes: controls, Alzheimer's disease, vascular dementia, frontotemporal lobar degeneration, and dementia with Lewy bodies.

Main Results:

• The Disease State Index classifier achieved a high balanced accuracy of 82.3% in separating the five classes.
• Classification accuracy was highest for vascular dementia and lowest for dementia with Lewy bodies.
• For the most confident 50% of classifications, the balanced accuracy reached 93.6%.

Conclusions:

• Data-driven CDSSs, like PredictND, can significantly aid in the differential diagnosis of dementias in clinical practice.
• The PredictND tool demonstrates high accuracy in distinguishing between different dementia subtypes and controls.
• The system's interpretability and confidence measure enhance its utility for medical decision support.