Histological tissue classification with a novel statistical filter-based convolutional neural network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Medical Imaging

Background:

• Convolutional Neural Networks (CNNs) are effective for image-based tasks but are computationally intensive.
• High-performing CNNs have numerous parameters, limiting their use on low-performance hardware.
• Existing CNNs face challenges in efficient feature extraction for complex image datasets.

Purpose of the Study:

• To introduce a novel statistical filter-based CNN (HistStatCNN) for improved image classification.
• To address the computational demands and parameter limitations of traditional CNNs.
• To enhance the accuracy and efficiency of deep learning models in histopathological image analysis.

Main Methods:

• Designed a CNN model with convolution kernels initialized using continuous statistical methods.
• Evaluated the HistStatCNN on a new histological dataset and several histopathological benchmark datasets.
• Applied unique and mixed parameter sets of statistical filters to the CNN for classification tasks.

Main Results:

• The proposed HistStatCNN achieved 87.13% accuracy for histological data classification, outperforming standard models like GoogleNet and ResNet variants.
• Tested on various histopathological datasets, the statistical filter initialization significantly increased average accuracy rates.
• Experimental results confirmed that statistical filters enhance CNN performance, even with simpler model architectures.

Conclusions:

• The novel statistical filter-based approach (HistStatCNN) effectively enhances CNN performance for image classification tasks.
• HistStatCNN offers a more computationally efficient and accurate solution for histopathological image analysis.
• The proposed filter initialization method demonstrates broad applicability and improved results across diverse datasets.