Classification of endometrial lesions by nuclear morphometry features extracted from liquid-based cytology samples: a system based on logistic regression model.

Objective: To investigate the potential of a computerized system for the discrimination of benign from malignant endometrial nuclei and lesions.

Study design: A total of 228 histologically confirmed liquid-based cytological smears were collected: 117 within normal limits cases, 66 malignant cases, 37 hyperplasias without atypia, and 8 cases of hyperplasia with atypia. From each case we extracted nuclear morphometric features from about 100 nuclei using a custom image analysis system. Initially we performed feature selection, and subsequently we applied a logistic regression model that classified each nucleus as benign or malignant. Based on the results of the nucleus classification process, we constructed an algorithm to discriminate endometrium cases as benign or malignant.

Results: The proposed system had an overall accuracy for the classification of endometrial nuclei equal to 83.02%, specificity of 85.09%, and sensitivity of 77.01%. For the case classification the overall accuracy was 92.98%, specificity was 92.86%, and sensitivity was 93.24%.

Conclusion: The proposed computerized system can be applied for the classification of endometrial nuclei and lesions as it outperformed the standard cytological diagnosis. This study highlights interesting diagnostic features of endometrial nuclear morphology, and the proposed method can be a useful tool in the everyday practice of the cytological laboratory.