Table 5.
Summary of publications focused on applying deep learning methods for prostate histopathology images.
Method | Reference | Year | Accuracy Result | Software | ||
---|---|---|---|---|---|---|
CNN | [31] | 2016 | AUC ranges from 0.88 to 0.99. | N/A | ||
CNN built upon VGG19 | [27] | 2019 | Average accuracy of classifying Artefacts vs. Glands is 95.4%, average accuracy of classifying Benign vs. Pathological is 88.3%, Average accuracy of Multi-class classification is 87.6% | Matlab 2018b + Python 3.5 with Keras library and Tensorflow as backend. | ||
Pretrained CNN | [30] | 2016 | The classification accuracy per image patch is 81%, while for the whole images, the classification accuracy is 89%. | N/A | ||
Different CNN Architectures | ResNet-50 | [28] | 2018 | They evaluated their results using test cohort and they observed that MobileNet attained the best performance on the validation set | Python 3 with Keras library and tensorflow as backend. Some analysis was done in R by the help of using survminer and survival packages. | |
MobileNet | ||||||
Inception-V3 | ||||||
DenseNet-121 | ||||||
VGG-16 | ||||||
U-Net | [29] | 2020 | The developed model achieved accuracy of 99% for biopsies containing tumor and a specificity of 82%. | Tensorflow and Keras | ||
SSA-RideNN | [23] | 2019 | The technique achieved maximal accuracy of 89.6% and sensitivity of 89.1%, and specificity of 85.9% | Matlab | ||
SVM | [34] | 2018 | They used Cohen’s kappa coefficient to evaluate the performance. The highest value attained is 0.52 by logistic regression, while 0.37 is attained by using CNN. | Matlab | ||
Random forest | ||||||
linear discriminant analysis | ||||||
logistic regression | ||||||
CNN | ||||||
Different CNN Architectures | EfficientNet | [113] | 2020 | UNet attained the best result of AUC about 0.98 | N/A | |
DenseNet | ||||||
U-Net | ||||||
cGAN | [25] | 2018 | The proposed technique achieved F1-score 85.7% for prostate dataset | Pytorch 0.4 | ||
NB that utilizes CNN | [26] | 2019 | Their proposed model achieves 81.3% precision, 91.4% in recall, and 85.4% in F1. | Python 2.7 with Keras library and Tensorflow | ||
Path RCNN | [24] | 2019 | Path RCNN attained accuracy of 99% and a mean of area under the curve of 0.99. | Python and Tensorflow backend |