Аннотации:
Breast cancer is the most common type of cancer, with over 2.2 million cases reported
in 2020. Breast cancer treatment can be highly effective, especially when the disease
can be detected at an early stage. Nowadays, scientists provide many solutions to
identify the type of tumor in the early stages through medical imaging. In this
study, our task is to use pre-trained deep CNN architectures to find the type of
tumor and develop these models using different optimization methods, changing the
parameters of settings for these models, and applying data augmentations methods
to the medical images which in turn, yield models with better accuracy. During
this work, we used six different models on five datasets from the same database and
improved the results through different data augmentation methods. A database of
histopathological images was used, where the data are divided into malignant, benign,
and pre-trained models were also used as Efficientnetv2, Mobilenet-v2, Resnetv2- 50,
VGG16, Inception-v3, and Inception-Resnet-v2 Almost all models performed well
after fine-tuning histopathological images, but EfficientNet showed the best result
with 94.5% accuracy. In addition, it performed well on magnified under a microscope
200x histopathological images and achieved 98% accuracy.