Skip to main content

Table 3 Details of Results of Experiments 1 and 2

From: A convolutional neural network-based system to classify patients using FDG PET/CT examinations

Experiment 1
(a) Image-based Correct Label
Benign Malignant Equivocal
Prediction Benign 96.6% 2.4% 10.1%
Malignant 0.3% 97.3% 12.1%
Equivocal 3.2% 0.2% 77.8%
(b) Image-based Evaluation Measures   Recall score Precision score F measure
Prediction Benign 0.966 0.917 0.941
Malignant 0.973 0.936 0.954
Equivocal 0.778 0.986 0.87
(c) Patient-based Algorithm A Correct Label
Benign Malignant Equivocal
Prediction Benign 91.0% 0.0% 0.0%
Malignant 9.0% 100.0% 42.5%
Equivocal 0.0% 0.0% 57.5%
(d) Patient -based Algorithm A Evaluation Measures   Recall score Precision score F measure
Prediction Benign 0.910 1.000 0.953
Malignant 1.000 0.764 0.866
Equivocal 0.575 1.000 0.730
(e) Patient-based Algorithm B Correct Label
Benign Malignant Equivocal
Prediction Benign 99.4% 0.6% 3.8%
Malignant 0.6% 99.4% 8.8%
Equivocal 0.0% 0.0% 87.5%
(f) Patient -based Algorithm B Evaluation Measures   Recall score Precision score F measure
Prediction Benign 0.994 0.975 0.984
Malignant 0.994 0.951 0.972
Equivocal 0.875 1.000 0.933
Experiment 2
(g) Head and Neck Correct Label
Benign Malignant Equivocal
Prediction Benign 97.8% 1.7% 3.0%
Malignant 1.5% 97.3% 0.8%
Equivocal 0.7% 1.1% 96.2%
(hd) Chest Correct Label
Benign Malignant Equivocal
Prediction Benign 98.4% 1.8% 5.9%
Malignant 0.6% 96.6% 1.6%
Equivocal 1.0% 1.6% 92.5%
(i) Abdomen Correct Label
Benign Malignant Equivocal
Prediction Benign 94.9% 5.7% 7.0%
Malignant 1.1% 92.8% 2.0%
Equivocal 4.1% 1.5% 91.0%
(j) Pelvic region Correct Label
Benign Malignant Equivocal
Prediction Benign 99.7% 0.4% 2.8%
Malignant 0.1% 99.6% 1.9%
Equivocal 0.3% 0.0% 95.3%