Skip to main content
. 2023 Jan 9;12:1044026. doi: 10.3389/fonc.2022.1044026

Table 2.

Comparison of different classification methods.

Normal Benign in situ carcinoma Invasive carcinoma
Model Accuracy P R P R P R P R
Supervised VGG (9) 56.3 ( ± 10.9) 60.0 60.0 61.1 55.0 60.0 60.0 62.5 45.5
Inception (26) 71.3 ( ± 9.9) 68.4 65.0 71.4 75.0 68.1 75.0 66.7 77.8
ResNet101 (44) 86.3 ( ± 7.5) 89.4 85.0 81.8 90.0 80.9 85.0 94.4 85.0
Semi-supervised Pseudo-Labeling (30) 61.3 ( ± 10.7) 57.9 55.0 61.9 65.0 52.4 55.0 73.7 70.0
Mean Teacher (48) 70.0 ( ± 10.0) 66.7 70.0 68.4 65.0 70.0 70.0 75.0 75.0
MixMatch (49) 87.5 ( ± 7.2) 90.0 90.0 80.9 85.0 85.0 85.0 94.7 90.0
FixMatch (14) 87.5 ( ± 7.2) 94.4 85.0 85.0 85.0 81.8 90.0 90.0 90.0
Semi-His-Net 90.0 (± 6.6) 90.0 90.0 89.4 85.0 85.7 90.0 95.0 95.0

We show the average Accuracy with a 95% confidence interval in parentheses, Precision (P) and recall (R) (%) of four subtypes: normal tissue, benign abnormality, malignant carcinoma in situ, and malignant invasive carcinoma. Bold font indicates best result obtained for predictions.