Table 1 Classification performance comparison of the architecture trained from the beginning with large pre-trained models in the test set

From: A convolutional attention model for predicting response to chemo-immunotherapy from ultrasound elastography in mouse tumor models

Model

AUC

Accuracy

Sensitivity

Specificity

PPV

NPV

Prognose-CNNattention

0.96 ± 0.0012

0.8683 ± 0.0071

0.8676 ± 0.0009

0.9349 ± 0.0012

0.8629 ± 0.0011

0.9355 ± 0.0022

Prognose-CNN (without Attention)

0.96 ± 0.0021

0.8491 ± 0.0082

0.8453 ± 0.0010

0.9257 ± 0.0001

0.8466 ± 0.0023

0.9108 ± 0.0034

Auto-Prognose-CNNattention

0.95 ± 0.0015

0.8299 ± 0.0113

0.8232 ± 0.0009

0.9164 ± 0.0002

0.8339 ± 0.0001

0.9152 ± 0.0043

Xception

0.92 ± 0.0023

0.8095 ± 0.0020

0.8027 ± 0.0009

0.9058 ± 0.0007

0.8126 ± 0.0014

0.9047 ± 0.0012

VGG16

0.94 ± 0.0014

0.7949 ± 0.0048

0.7855 ± 0.0011

0.8983 ± 0.0007

0.7977 ± 0.0011

0.8994 ± 0.0023

Inception-V3

0.91 ± 0.0012

0.7875 ± 0.0092

0.7582 ± 0.0002

0.8848 ± 0.0009

0.7603 ± 0.0031

0.8909 ± 0.0026

ResNet50

0.92 ± 0.0011

0.8172 ± 0.0014

0.8127 ± 0.0001

0.9079 ± 0.0011

0.8137 ± 0.0021

0.9075 ± 0.0025

  1. Macro-average values of AUC, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) are presented for each model. Macro-average is the average value of each metric for the three classes, by treating all 3 classes equally. All pre-trained models were fine-tuned to the specific task and data.
Back to article page