Joint multiple fully connected convolutional neural network with extreme learning machine for hepatocellular...

Area of Science:

• Medical Image Analysis
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Accurate grading of cancerous tissue pathological images is crucial for effective medical diagnosis and treatment planning.
• Hepatocellular carcinoma (HCC) nuclei grading requires precise analysis of histopathological features.
• Existing methods may face challenges in capturing complex multi-scale contextual information for accurate grading.

Purpose of the Study:

• To propose a novel joint multiple fully connected convolutional neural network with extreme learning machine (MFC-CNN-ELM) architecture for automated HCC nuclei grading.
• To enhance the accuracy and generalization capability of pathological image analysis for cancer diagnosis.

Main Methods:

• Utilized center-proliferation segmentation (CPS) for preprocessing grayscale image patches, with labels guided by expert pathologists.
• Developed a multiple fully connected convolutional neural network (MFC-CNN) to automatically extract multi-form feature vectors, considering multi-scale contextual information.
• Employed a convolutional neural network extreme learning machine (CNN-ELM) model for HCC nuclei grading, trained using a back propagation (BP) algorithm with a novel up-sample method.

Main Results:

• The proposed MFC-CNN-ELM architecture achieved superior performance in HCC nuclei grading compared to existing methods.
• Experimental comparisons validated the effectiveness of the MFC-CNN-ELM approach.
• External validation on the ICPR 2014 HEp-2 cell dataset confirmed the good generalization ability of the proposed architecture.

Conclusions:

• The MFC-CNN-ELM architecture offers a robust and accurate solution for hepatocellular carcinoma nuclei grading.
• The study highlights the potential of deep learning, specifically MFC-CNN-ELM, in advancing computational pathology and cancer diagnosis.
• The developed method demonstrates significant improvements in accuracy and generalization for pathological image analysis.