Neuro-VisAge

The last decade was marked by a spike in the use of Machine Learning (ML) and functional magnetic resonance image for neurological disorders diagnoses. However, few publications consider age bias when designing their experiments, which can impact the final result. Here, we investigate the effects of age bias on a sample of typical neurological subjects, looking for patterns in brain activity. We also suggest that age groups be used in the ML training and classification for future works. Our results show that for the five brain regions investigated (Frontal Gyrus, Cingulum Bundle, Putamen, Angular Gyrus, and Heschl Gyrus), using a 10-year span would increase the reliability of ML experiments aiming to diagnose neural disorders by reducing the age bias effect on the models.

Equations

The raw sample on the Neuro-VisAge comprises the tuple average and standard deviation of each ROI of each subject. We grouped the values of each age. Therefore, Neuro-VisAge shows two averages (Equations 1 and 2) and two standard deviations (Equations 3 and 4).

For all equations, a represents a set of subjects of the same age, ranging from six to 56, and Xj represents all values of an ROI from a subject; therefore, the AA from Equation 1 represents the average for age a of the functional activity average of all subjects of age a for the ROI j. Equation 2 represents the average for age a of the functional activity standard deviation of all subjects of age a for the ROI j. Equation 3 represents the standard deviation for age a of the functional activity average of all subjects of age a for the ROI j. Equation 4 represents the standard deviation for age a of the functional activity standard deviation of all subjects of age a for the ROI j. Finally, Equations 5 and 6 represent the average and standard deviation analog from the other, but they group the subjects by FIQ instead of age.

Learn more

Neuro-VisAge

Equations

Source code

Implementation

Demographics

Original DataSet

Authors