Semi-automatic segmentation and classification of Pap smear cells.

Area of Science:

• Biomedical Engineering
• Computational Pathology
• Cytopathology

Background:

• Cervical cancer screening relies heavily on cytologic analysis.
• Automated and semiautomated systems can enhance the efficiency and accuracy of cell analysis.

Purpose of the Study:

• To develop a semiautomatic PC-based cellular image analysis system for cervical cancer screening.
• To classify different cell types and discriminate between normal and dysplastic cells.

Main Methods:

• Developed a system for segmenting nuclear and cytoplasmic contours.
• Extracted morphometric and textual features for Support Vector Machine (SVM) classification.
• Implemented a software program for standardized workflow and image reviewing.

Main Results:

• Achieved high accuracies in cross-validation: 97.16% for four cell types and 98.83% for dysplastic vs. normal cells.
• In a separate experiment, accuracies reached 96.12% (four-cluster) and 98.61% (two-cluster) with 70% training and 30% testing data.
• SVM recursive feature addition was used for feature selection.

Conclusions:

• The developed semiautomatic system is a feasible and effective tool for evaluating cytologic specimens.
• The system demonstrates significant potential for improving cervical cancer detection accuracy and workflow standardization.