Table 8 Examples of the relationship between features and class prediction.

From: A multilayer multimodal detection and prediction model based on explainable artificial intelligence for Alzheimer’s disease

First Layer

Most influential feature

CDRSB

Clinical dementia rating sum of box score

 

Lowest important feature

TRABSCOR PartBTimeToComplete

Neuropsychological battery’s TRABSCOR trail making test (part B—time to complete)

The best feature for AD class

MMSE

Mini-Mental State Examination

The best feature for CN and MCI classes

CDRSB

Clinical dementia rating sum of box score

\(\uparrow\) CDRSB, \(\downarrow\) ADNI_MEM

\(\uparrow\) risk for the AD class

Sensitivity of the AD class to this list of features

\(\downarrow\) CDRSB, \(\uparrow\) ADNI_MEM, \(\uparrow\) DigitalTotalScore, \(\uparrow\) MOCA

\(\uparrow\) chance for the CN class

Sensitivity of the CN class to this list of features

\(\downarrow\) CDRSB, \(\uparrow\) FAQ,\(\uparrow\) MOCA, \(\uparrow\) CDGLOBAL, \(\downarrow\) ADNI_MEM

\(\downarrow\) risk for the MCI class

Sensitivity of the MCI class to this list of features

Second Layer

Most influential feature

ADNI_MEM

ADNI_MEM is composite logical memory score for the8

longitudinal changes in memory

 

Lowest important feature

Trail4Total

Neuropsychological Battery AVTOT4 feature

\(\uparrow\) FAQ, \(\uparrow\) RAVLT_immediate

\(\uparrow\) chance for the sMCI class

Relationship between RAVLT_immediate and sMCI class

\(\uparrow\) FAQ, \(\uparrow\) RAVLT_immediate

\(\downarrow\) risk for the pMCI class

Relationship between FAQ and RAVLT_immediate and pMCI class

\(\uparrow\) ADAS 13, \(\uparrow\) ADNI_MEM, \(\downarrow\) FDG, \(\downarrow\) MOCA

\(\uparrow\) risk for the pMCI class

Relationship between ADAS, ADNI_MEM, FDG, and MOCA and pMCI class