Assessing the Impact of Data Augmentation on Transfer Learning-Based Medical Image Classification

Statistics augmentation is making extra categorized information factors from present datasets. This system has been hired to improve the performance of deep learning-primarily based scientific picture classification duties, even though its impact on transfer gaining knowledge of-primarily based medical photograph category obligations stay unknown. Subsequently, this paper assesses the effect of information augmentation on switch-getting to know-based medical photograph category responsibilities. We retrospectively evaluated three different architectures pre-trained on the Image Net dataset, namely Resnet18, Resnet50, and InceptionV3, on our CXR-14 dataset, including posterior-anterior chest X-ray pics. We determined that data augmentation increased the classification accuracy using 4% on ResNet18, 6% on ResNet50, and 2.5% on InceptionV3. Our take look proves that applying statistics augmentation to switch learning-primarily based clinical photograph classification responsibilities improves accuracy, which was no longer previously acknowledged. We envisage that this finding may have a crucial bearing on the scientific imaging subject by presenting an extra effective way of leveraging current datasets to enhance medical picture class duties.