Efficient Diagnosis of Cancer from Histopathological Images By Eliminating Batch Effects

Abstract

Computer aided decision support systems has been developed by researchers for efficiently diagnosing cancer from histopathological images. The presence of batch effects in histopathological images may reduce the prediction performance of these systems, especially when photos are obtained making use of diverse image resolution devices and patient samples. It is even more bothersome with large-scale variations in which cross laboratory sharing of large volumes of data is necessary. Batch effects can alter quantitative morphological image features and also decreases the actual conjecture performance. Principal component variation analysis ensures that batch effects may cause large variations in image features. Using the four batches of renal tumor images, one image-level and four feature-level batch effect removal methods are compared. Among this combat normalisation method performance is much better than the performance of other batch effect removal methods.