Table 1 Detection and segmentation performance of the proposed multi-task model vs. single-task baselines are tested on the above dataset, shown in Table 8

From: GLANCE: continuous global-local exchange with consensus fusion for robust nodule segmentation

Model variant

Detection performance

Segmentation

 

Sensitivity @1 FP

Sensitivity @4 FP

CPM (avg. sens)

Dice

Detection-only

80.2

92.5

90.1

–

Segmentation-only

–

–

–

95.2

Ours (multi-task model)

95.7

94.8

92.1

96.4

  1. Sensitivity is reported at 1 and 4 false positives per scan (FP/s), along with the competition performance metric (CPM). Segmentation accuracy is reported as the Dice coefficient.
Back to article page