Machine learning for classification of cutaneous sebaceous neoplasms: implementing decision tree model using...

Area of Science:

• Dermatopathology
• Computational pathology
• Oncology

Background:

• Sebaceous lesions present diagnostic challenges due to overlapping histopathological features.
• Accurate classification is crucial for appropriate patient management and treatment.

Purpose of the Study:

• To describe and compare histopathological, architectural, and nuclear characteristics of sebaceous lesions.
• To develop a predictive classification model for sebaceous lesions using machine learning.

Main Methods:

• Cross-sectional study of 123 Iranian patients with sebaceous tumors (March 2015-March 2019).
• Pathology slides reviewed for architectural and cytological attributes.
• Multiple decision tree models employed with 5-fold cross-validation.

Main Results:

• Histopathological and nuclear features like pagetoid appearance, irregular nuclear contour, and large nuclear size were exclusive to carcinomatous tumors.
• Benign lesions (sebaceoma, adenoma) showed potentially misleading features such as high mitotic activity.
• Key predictors for classification: basaloid cell count, peripheral basaloid cell layers, tumor margin, nuclear size, and chromatin.

Conclusions:

• Machine learning effectively categorizes sebaceous lesions based on architectural and nuclear features.
• The developed predictive model aids in differentiating benign from malignant sebaceous neoplasms.
• Further validation with larger sample sizes is recommended to confirm the model's accuracy.