Refinement of Convolutional Neural Network Based Cell Nuclei Detection Using Bayesian Inference.

Area of Science:

• Computational pathology
• Medical image analysis
• Cancer diagnostics

Background:

• Manual microscopic analysis of cytological samples for cancer diagnostics is time-consuming and prone to inter-observer variability.
• Analyzing clumped cellular structures in cytological samples poses challenges for accurate morphometric parameter extraction.
• Automated tools are needed to assist pathologists in objective and efficient analysis of cellular structures.

Purpose of the Study:

• To develop and evaluate an automated nuclei detection approach for cytological samples.
• To address the challenge of analyzing clumped and occluded cell nuclei.
• To improve the accuracy and efficiency of cancer diagnostics through computational methods.

Main Methods:

• A novel approach combining convolutional neural networks (CNNs) for semantic segmentation and Bayesian inference using marked point processes.
• Preprocessing involved stain separation to isolate hematoxylin, highlighting cell nuclei.
• Post-processing utilized Besag's iterated conditional modes to model nuclei distribution and fit overlapping ellipses to represent clusters.

Main Results:

• The proposed method achieved high accuracy in nuclei detection on 50 breast cancer cytological images.
• Achieved 93.5% true positive (TP) nuclei detection rate with only 6.1% false positive (FP) rate.
• Outperformed the marker-controlled watershed method in both correctly detected nuclei and reduced false detections.

Conclusions:

• The combined CNN and Bayesian inference approach effectively detects cell nuclei in challenging cytological samples.
• This method offers a significant improvement over traditional techniques like marker-controlled watershed for automated nuclei detection.
• The developed tool has the potential to enhance the accuracy and efficiency of cancer diagnosis by assisting pathologists.