Pooling in convolutional neural networks for medical image analysis: a survey and an empirical study

Abstract

Convolutional neural networks (CNN) are widely used in computer vision and medical image analysis as the state-of-the-art technique. In CNN, pooling layers are included mainly for downsampling the feature maps by aggregating features from local regions. Pooling can help CNN to learn invariant features and reduce computational complexity. Although the max and the average pooling are the widely used ones, various other pooling techniques are also proposed for different purposes, which include techniques to reduce overfitting, to capture higher-order information such as correlation between features, to capture spatial or structural information, etc. As not all of these pooling techniques are well-explored for medical image analysis, this paper provides a comprehensive review of various pooling techniques proposed in the literature of computer vision and medical image analysis. In addition, an extensive set of experiments are conducted to compare a selected set of pooling techniques on two different medical image classification problems, namely HEp-2 cells and diabetic retinopathy image classification. Experiments suggest that the most appropriate pooling mechanism for a particular classification task is related to the scale of the class-specific features with respect to the image size. As this is the first work focusing on pooling techniques for the application of medical image analysis, we believe that this review and the comparative study will provide a guideline to the choice of pooling mechanisms for various medical image analysis tasks. In addition, by carefully choosing the pooling operations with the standard ResNet architecture, we show new state-of-the-art results on both HEp-2 cells and diabetic retinopathy image datasets.

A survey on deep learning applied to medical images: from simple artificial neural networks to generative models

“…The most used variants are: max or min pooling (higher or lower value goes to the input unmodified) and average pooling (an average value of all the input values affected is obtained as output). Nirthika et al [ 49 ] perform an empirical study of pooling operations in CNN for medical image analysis. They conclude that choosing an appropriate pooling technique for a particular job is related to the size and scale of the images and its class-specific features.…”

Fusing compressed deep ConvNets with a self-normalizing residual block and alpha dropout for a cost-efficient classification and diagnosis of gastrointestinal tract diseases

“…In this paper, the selected output shape became 7 2 × 192, as the filter size f of 192 provides better adequacy than 96, but still less than 512, as the proposed method considers the importance of cost. By providing the specific kernel size k , strides s , pool size p, and other settings after the cut-points of each model, they eventually had a similar output shape of 7 2 × 192, making them compatible for a layer-wise fusion [21] . On the other hand, the auxiliary layers also contain an alpha DO with a rate of 0.2 that improves the regularization of the incoming fused features but does not affect the reshaping of each model's output.…”

“…On the other hand, the auxiliary layers also contain an alpha DO with a rate of 0.2 that improves the regularization of the incoming fused features but does not affect the reshaping of each model's output. If cases that other researchers or users use cut-points that do not reflect the ones in this proposed method, the Conv and AP settings will necessitate alterations to make them fit [21] .…”

Deep learning model coupled with proteomics to detect yeast epidemic clones across multiple hospitals

“…7 is a very simple convolutional neural network (CNN) [23] model taking a spectrum of 18,000 values as input. The convolutional layer was made of 3 lters and a kernel size of 6, followed by a max pooling layer with pool size = 100 [24]. The atten output of the CNN was sent to two fully connected layers of respectively 512 and 1024 units.…”