Artificial neural network analysis for evaluating cancer risk in multinodular goiter

Area of Science:

• Endocrinology
• Oncology
• Medical Informatics

Background:

• Multinodular goiter (MNG) presents diagnostic challenges, particularly with indeterminate cytology from fine-needle aspiration biopsies.
• Accurate differentiation between benign and malignant lesions in MNG is crucial for appropriate patient management.

Purpose of the Study:

• To develop and evaluate an artificial neural network (ANN) model for predicting malignancy in MNG patients with indeterminate cytology.
• To assess the diagnostic performance of the ANN model in distinguishing malignant from benign thyroid nodules.

Main Methods:

• Retrospective analysis of 116 MNG patients with indeterminate cytology.
• Collection of clinical data including age, sex, nodule size, thyroid hormone levels, and therapy.
• Development and validation of an ANN model using collected patient data.

Main Results:

• The study included 116 patients (89.7% female, mean age 44.6 years).
• Final pathology confirmed 20.7% malignant and 79.3% benign diseases.
• The ANN model achieved a high accuracy in predicting malignancy, with an area under the curve (AUC) of 0.824.

Conclusions:

• Artificial neural networks (ANNs) show promise as a valuable tool to aid in the diagnosis of malignancy in MNG.
• The ANN technique may help decrease the rate of unnecessary thyroidectomies in patients with indeterminate thyroid cytology.
• Further research is warranted to refine the diagnostic model and integrate it into clinical practice.