Selecting the optimal transfer learning model for precise breast cancer diagnosis utilizing pre-trained deep learning models and histopathology images

BackgroundEvery year, around 1.5 million women worldwide receive a breast cancer diagnosis, which is why the mortality rate for women is rising. Scientists have developed Convolutional neural network models in recent years to simplify the breast cancer diagnosis process. CNN displays encouraging findings for cancer classification using image datasets. However, there are not yet any best-in-class standard models because big datasets are insufficient for training and verifying models.MethodTo fully utilize the transfer learning technique, researchers are now focusing on using pre-trained feature extraction models trained on billions of images. The parallel processing of data at several clusters is necessary to keep up with the continually expanding datasets. Two factors are necessary to design a perfect and precise breast cancer diagnostic neural model. One is that the selected imaging modality will primarily determine the model's prediction path. The neural network model employed for breast cancer prediction and the environment in which it is applied comes in second. Data processing in parallel across several clusters and hardware demands for greater processing capacity, such as GPU and TPU, is necessary to keep up with the continuously expanding datasets.ResultsThe impact of high-performance computing and a critical examination of the pre-trained models employed in breast cancer picture categorization are discussed. In this paper, a thorough analysis of image modality's influence on the accuracy of breast cancer detection is done. In addition, the primary breast cancer detection pre-trained models are reviewed, and the effects of HPC on CNN training are investigated. Major transfer learning techniques (VGG16, Xception, and others) are used in the case study to analyze an image collection of invasive ductal carcinoma (IDC), a type of breast cancer. The study suggests using CNN architectures built on deep neural networks and pre-trained networks to identify breast cancer. The results show that CNN performs better for feature extraction, optimization, and classification of breast cancer images when GPU is used. The length of training has drastically changed. High-powered systems speed up the process of identifying cancer, which facilitates the analysis of patterns in the dataset-something that is impossible for humans to do. High-performance technologies make it simple to analyze complicated patterns.