Table 4.

Validation of slice-level classification results using multi-tasks learning with four backbone CNN architectures (ResneXt-50, Densenet-161, Inception-v3, and Wide-Resnet-50) [6].

Model	Slice-Level Classification (%)
ResneXt-50	88.91
Densenet-161	88.61
Inception-v3	88.74
Wide-Resnet-50	88.37