Breast cancer mitotic cell detection using cascade convolutional neural network with U-Net.

Area of Science:

• Digital pathology
• Computational oncology
• Medical image analysis

Background:

• Accurate mitotic cell counting is crucial for breast cancer prognosis but remains challenging for pathologists.
• Traditional methods using feature extraction and sliding windows are often time-consuming and ineffective on high-resolution pathological images with complex backgrounds.

Purpose of the Study:

• To develop an automated, efficient, and accurate method for detecting mitotic tumor cells in breast cancer pathology slides.
• To address the limitations of traditional manual and deep learning-based approaches in mitotic cell detection.

Main Methods:

• Proposed a cascaded convolutional neural network (CNN) named UBCNN, integrating semantic segmentation and classification.
• Employed an improved UNet++ for initial target localization and an improved 2D VNet for precise cell nucleus segmentation.
• Utilized a trained CNN for binary classification of segmented cell nuclei to identify mitotic cells.

Main Results:

• The UBCNN algorithm achieved an accuracy of 0.831 on the ICPR 2012 dataset and 0.576 on the ICPR 2014 dataset.
• Demonstrated improved performance in terms of accuracy, recall, and F-score compared to existing algorithms.
• The cascaded approach effectively screened candidate areas to retain final mitotic cell detections.

Conclusions:

• The proposed UBCNN cascaded detection algorithm offers a competitive and improved solution for automated mitotic cell detection in breast cancer.
• This method enhances efficiency and accuracy in a critical aspect of cancer prognosis, aiding pathologists.
• The integration of semantic segmentation and classification within a cascaded CNN framework shows significant promise for digital pathology applications.