The proposed study aims to develop and validate robust diagnostic protocols for neurodegenerative conditions based on quantitative imaging measures of disease-specific pathognomonic brain regions.
Patients with Amyotrophic Lateral Sclerosis, patients with other neurodegenerative conditionsand healthy controls will be dividedinto a ‘training cohort’ for classification model development and an independent validation sample to evaluate the diagnostic accuracy of the models. Comparative analyses between controls and ALS patients will be undertaken in the training cohort to identify key, disease-specific imaging signatures. As the analyses take place in a standardized radiological reference system (MNI), anatomical coordinates for these discriminative regions will be recorded as MNI-defined “disease-defining” regions of interest (ROI). Key, unaffected, “disease-defying” brain regions in ALS will be utilised to identify discriminatory regions from mimic neurodegenerative conditions. Normative MRI values of these ROIs will be determined from healthy control data in the training sample.
In the validation step, blinded data sets will also be registered to standard MNI space and virtual “data biopsies” will be taken from the previously identified “disease-defining” and “disease-defying”ROIs. Based on data retrieved from these ROIs, participants will be classified as ALS, healthy control or mimic neurodegenerative condition using binary logistic regression, random forest approaches, naive Bayes classification, and support vector learning machines. The novelty of the project is the inclusion of disease-defiant brain regions, inclusion of mimic controls, integration of multiple imaging indices of multiple anatomical regions.
The accuracy, sensitivity, and specificity of the four classification methods will be evaluated both in the training cohort and in the independent validation cohort. This approach enables the critical appraisal of the discriminatory power of the four most commonly used classification models to identify which approach yields to optimal diagnostic accuracy and has the potential to be adapted for clinical use.
