Kidney Whole Slide Image Quality Assessment Based on Fused Convolutional Neural Networks

The quality of the renal biopsy sections is very important in diagnosing diffuse renal disease. However, there is currently no standard to assess the quality of the image due to the subjectivity of the doctor. Therefore, this paper proposes a novel no-reference quality assessment method for kidney pathological whole slide images (WSI), which introduced a deep neural network in this task for the first time. In this method, the multiple CNN architectures are fused and jointly trained to capture the characteristics of pathological images better, and the input images are automatically classified into four levels: excellent, good, average, and poor. Our model is evaluated in a quality assessment database which collected and annotated via three senior pathologists. As revealed by the experiment, our method exceeds the previous CNN architecture by reaching an accuracy of 0.851. Our work establishes the standard for the assessment of pathological section images, which not only realizes the automatic assessment of the quality for kidney pathological whole slide images, but also can even provide guidance for the standardization and restoration of sections in the future.