Computerized diagnosis of prostate cancer based on whole slide histology images and deep learning methods
Вантажиться...
Дата
2021
Автори
Назва журналу
Номер ISSN
Назва тому
Видавець
Анотація
This paper addresses the problem of the diagnosis of prostate cancer
and providing the final score in the commonly used ISUP scale. The input data
consisted of 10 616 whole-slide histological images with the size of the largest
dimension up to 100 000 pixels and 22 089 of their image tiles of 256x256 pixels in
size. Several approaches were examined including conventional methods based on
feature extraction followed by suitable classifiers as well as various deep learning
techniques based on convolutional neural networks of different architectures. During
the experiments, two deep learning approaches have been shown to be the most
efficient. The first one implements a technique of sequential, tile-by-tile image
classification using ResNet50 as a feature extractor followed by a prediction of final
ISUP score based on the informative features using a small convolutional network. The
second approach predicts the final ISUP score using an ensemble of 4 deep learning
models for regression from a subset of the most prominent image tiles first, combined
into one single pseudo-image. Being independently tested on an unknown image
dataset that was not available for authors, these approaches achieved the prediction
accuracy of 0.8110 and 0.9277 respectively.
Опис
Malyshev V. Computerized diagnosis of prostate cancer based on whole slide histology images and deep learning methods / V. Malyshev, D. Voynov, E. Lapo ; advisor : V. Kovalev // Black Sea Science 2021 : proc. of the Intern. Competition of Student Scientific Works / Odessa National Academy of Food Technologies ; eds. B. Yegorov, M. Mardar [et al.]. – Odessa: ONAFT, 2021. – P. 391–401 : tabl., fig. – Ref.: 11 tit.
Ключові слова
prostate cancer, whole slide histology, deep learning, convolutional neural networks