Histopathological image classification using CNN with squeeze and excitation networks based on hybrid squeezing

Histopathological image analysis of biopsy sample is the most reliable method for the detection and diagnosis of cancer. Automation in histopathological image analysis will help the pathologists to confirm their remarks with a second judgment. The proposed framework employs a CNN model with squeeze and excitation (SE) module based on hybrid squeezing method. In this approach, two levels of squeezing are provided for the feature maps using color-based spatial squeezing and channel-wise pooling. This squeezed weight adaptively scales each channel by boosting meaningful feature maps and diminishing less important features. The proposed CNN model is tested for the classification of histopathological images using Camelyon 16 and BreaKHis dataset. The experiments were conducted in four phases such as (i) CNN model without squeeze and excitation module (ii) CNN model with only channel pooling method (iii) CNN model with color-based spatial squeezing method (iv) CNN model with color-based spatial squeezing and channel pooling SE block. From the experimental results, the proposed model confirms better performance for histopathological image classification in terms of accuracy, precision, recall, F1 score and ROC. The computational load of the proposed model is also evaluated against regular CNN without SENet for obtaining the same evaluation metrics. The result shows the proposed model contributes 35% reduction in computational load in terms of trainable parameters. The performance of the proposed model is compared with state-of-the-art CNN methods and it is proved that the proposed model outperforms well in terms of evaluation metrics with very few numbers of model parameters.