Skip to main content

Table 5 Performance comparison of models using clinical data, positron emission tomography (PET) data, or dual modality

From: Deep learning model integrating positron emission tomography and clinical data for prognosis prediction in non-small cell lung cancer patients

Data

Model

MAE (days)

C-index

Classification accuracy of 2-year survival status

Classification accuracy of 5-year survival status

Clinical data

CPH

583 ± 37

0.747 ± 0.01

0.610 ± 0.05

0.868 ± 0.02

MLP {64 × 64}

463 ± 81

0.745 ± 0.01

0.740 ± 0.02

0.913 ± 0.03

PET (MIP images)

ResNet-50

423 ± 22

0.717 ± 0.01

0.724 ± 0.03

0.924 ± 0.01

PET (whole-body axial images)

ResNet3D-34

405 ± 29

0.749 ± 0.02

0.751 ± 0.02

0.928 ± 0.01

Clinical data + PET (whole-body axial images)

Multimodal

399 ± 27

0.756 ± 0.01

0.743 ± 0.02

0.933 ± 0.01

  1. The best score in each column is highlighted in bold
  2. MAE Mean absolute error; C-index Harrell’s concordance index; CPH Cox proportional hazards; MLP Multilayer perceptron; MIP Maximum intensity projection