Table 1 AUC performance of unsupervised anomaly detection on T1 scans using average \(\ell _2\) loss (among whole slice sets/continuous 10 slice sets exhibiting the highest loss) per scan. Unchanged CDR = 0 (i.e., cognitively healthy population) is compared against: (i) all the other CDRs (i.e., dementia); (ii) CDR = 0.5 (i.e., very mild dementia); (iii) CDR = 1 (i.e., mild dementia); (iv) CDR = 2 (i.e., moderate dementia). Each model is trained for 1.8M steps

MADGAN

0.768

0.750

0.797

0.829

MADGAN (10 slice sets)

0.764

0.745

0.793

0.830

MADGAN w/o \(\ell _1\) Loss

0.693

0.689

0.699

0.711

MADGAN w/o \(\ell _1\) Loss (10 slice sets)

0.705

0.697

0.717

0.736

3-SA MADGAN

0.752

0.736

0.775

0.835

3-SA MADGAN (10 slice sets)

0.739

0.725

0.760

0.810

3-SA MADGAN w/o \(\ell _1\) Loss

0.728

0.715

0.748

0.785

3-SA MADGAN w/o \(\ell _1\) Loss (10 slice sets)

0.735

0.721

0.756

0.806

7-SA MADGAN

0.765

0.743

0.800

0.832

7-SA MADGAN (10 slice sets)

0.764

0.743

0.798

0.835

7-SA MADGAN w/o \(\ell _1\) Loss

0.759

0.727

0.809

0.894

7-SA MADGAN w/o \(\ell _1\) Loss (10 slice sets)

0.746

0.710

0.803

0.868