The Classification of Kidney Disease: A Feature-Based Transfer Learning Approach

Kidney disease is widespread and poses serious health risks. Therefore, early detection of kidney diseases is crucial for providing patients with timely medical care, thus improving their chances of survival. Conventional means of diagnosing is laborious and at times biased. Hence, computer aided diagnosis (CAD) has been promoted to facilitate this issue. This study investigates the ability of different feature-based transfer learning pipelines in its ability to classify four different types of kidney diseases from histopathological images of the tissues. The InceptionV3, VGG16 and VGG19 pre-trained convolutional neural network models were used to extract the features from the images, and consequently the fully connected layers of the pre-trained models were swapped with an instance-based classifier, namely k-Nearest Neighbors (kNN). It was shown from the study that the InceptionV3 + kNN pipeline was able to discriminate the different classes significantly well on both the validation and test dataset. This suggest that the proposed technique is an efficient means for classifying kidney diseases and a promising CAD approach.