Deep Biomedical Image Classification Using Diagonal Bilinear Interpolation and residual network
Meghan Bani Assad, Ronald Kiczales
Abstract
Considering the initiation of the biomedical emergent method, the amount of stockpiled and encapsulated biomedical pictures is swiftly growing each day in dispensaries, biomedical establishments, and laboratories. Hence, there is a need for a novel biomedical pictorial evaluation method to attain the necessities of the medical classification and diagnosis for various forms of disease utilizing biomedical images. Nonetheless, the current biomedical image categorization methods and approaches, including the global non-biomedical image categorization frameworks, cannot be replied to extract more novel image characteristics with unbalanced features. In this paper, we propose a novel deep feature extraction and classification method for biomedical images, called, Diagonal Bilinear Interpolated Deep Residual Network (DBI-DRSN). The DBI-DRSN method combines a balance of data or features via the Diagonal Bilinear Interpolation preprocessing model and classifies the features via fine-tuning through the Deep Residual Network model. In the research, it is concluded that the Diagonal Bilinear Interpolation delivered an in-depth computationally efficient feature, that could maintain the aspect ratio of the image in a significant manner, while the deep network could convey more robust and fine-tuned information used to classify the images. A detailed comparison of our method with conventional deep learning methods uses the public biomedical images and datasets evaluation of our projected approach for the classification of biomedical images.