Microscopic nuclei classification, segmentation, and detection with improved deep convolutional neural networks...

Area of Science:

• Computational pathology
• Medical image analysis
• Deep learning in oncology

Background:

• Nuclei classification, segmentation, and detection in Whole Slide Images (WSI) are complex due to cellular heterogeneity.
• Accurate analysis of WSI is crucial for pathological diagnosis and cancer research.

Purpose of the Study:

• To propose advanced Deep Convolutional Neural Network (DCNN) models for nuclei classification, segmentation, and detection.
• To evaluate the performance of DCNN and Recurrent Residual U-Net (R2U-Net) based models on public datasets.

Main Methods:

• Applied Densely Connected Neural Network (DCNN) and Densely Connected Recurrent Convolutional Network (DCRN) for nuclei classification.
• Utilized Recurrent Residual U-Net (R2U-Net) and University of Dayton Net (UD-Net) for nuclei segmentation and detection.
• Conducted experiments on Routine Colon Cancer (RCC) and Nuclei Segmentation Challenge 2018 datasets, evaluated using five-fold cross-validation.

Main Results:

• Achieved 2.6% and 1.7% higher F1-score for nuclei classification and detection compared to existing DCNN methods.
• R2U-Net demonstrated 91.90% average testing accuracy (Dice Coefficient) for nuclei segmentation, surpassing U-Net by 1.54%.

Conclusions:

• The proposed DCNN and R2U-Net based methods exhibit robustness and superior performance in nuclei analysis tasks.
• These advanced models offer improved quantitative and qualitative results for Whole Slide Image (WSI) analysis.