Skip to main content

Table 2 AUC performance of unsupervised anomaly detection on T1c scans using average \(\ell _2\) loss (among whole slice sets/continuous 10 slice sets exhibiting the highest loss) per scan. No abnormal findings are compared against: (i) brain metastases + various diseases; (ii) brain metastases; (iii) various diseases. Each model is trained for 1.8M steps

From: MADGAN: unsupervised medical anomaly detection GAN using multiple adjacent brain MRI slice reconstruction

Normal versus BM + VD BM VD
MADGAN 0.765 0.888 0.618
MADGAN (10 slice sets) 0.769 0.905 0.607
MADGAN w/o \(\ell _1\) Loss 0.688 0.773 0.586
MADGAN w/o \(\ell _1\) Loss (10 slice sets) 0.696 0.778 0.597
3-SA MADGAN 0.756 0.859 0.633
3-SA MADGAN (10 slice sets) 0.760 0.871 0.626
3-SA MADGAN w/o \(\ell _1\) Loss 0.677 0.749 0.589
3-SA MADGAN w/o \(\ell _1\) Loss (10 slice sets) 0.708 0.780 0.622
7-SA MADGAN 0.781 0.921 0.613
7-SA MADGAN (10 slice sets) 0.776 0.917 0.608
7-SA MADGAN w/o \(\ell _1\) Loss 0.233 0.063 0.436
7-SA MADGAN w/o \(\ell _1\) Loss (10 slice sets) 0.234 0.091 0.405