Image processing for mitoses in sections of breast cancer: a feasibility study

Area of Science:

• Digital Pathology
• Computational Biology
• Oncology Research

Background:

• Accurate counting of mitotic nuclei is crucial for cancer diagnosis and prognosis.
• Manual scoring of mitoses by pathologists can be subjective and time-consuming.
• Automated image analysis offers potential for objective and efficient mitotic cell detection.

Purpose of the Study:

• To develop and evaluate an automated image analysis technique for counting mitotic nuclei in breast tissue sections.
• To assess the accuracy and efficiency of the proposed method compared to manual scoring by pathologists.

Main Methods:

• Image segmentation of tissue sections.
• Feature extraction using brightness histograms to reduce non-mitotic nuclei.
• Automated classification of remaining objects using contour features.
• Interactive evaluation of results with pathologists.

Main Results:

• The image analysis procedure aimed for a low false-negative rate but resulted in a 37% loss of mitotic nuclei.
• 85% of non-mitotic nuclei were eliminated, but 5% remained misclassified.
• Interactive evaluation showed 10% of mitotic nuclei were missed by the automated procedure.

Conclusions:

• The fully automatic image analysis procedure for mitotic nuclei counting in breast tissue sections yielded disappointing results.
• Further refinement is needed to improve the accuracy and reduce the misclassification rates of mitotic and non-mitotic nuclei.
• The current method requires significant improvement before clinical application in digital pathology.