Empirical Method for Thyroid Disease Classification Using a Machine Learning Approach.

Area of Science:

• Medical imaging analysis
• Computational biology
• Endocrinology

Background:

• Thyroid diseases, including hypothyroidism, hyperthyroidism, and thyroid cancer, affect global populations, causing severe symptoms.
• Timely detection and classification of thyroid diseases are crucial for effective patient treatment.
• Radiologists face challenges in accurate and efficient thyroid nodule detection from ultrasound images.

Purpose of the Study:

• To compare the effectiveness of various machine learning algorithms for thyroid disease classification.
• To evaluate the performance of algorithms on manipulated datasets for improved prediction accuracy.
• To identify the optimal machine learning model for precise thyroid nodule detection in ultrasound images.

Main Methods:

• Comparative analysis of machine learning algorithms: decision tree, random forest, k-nearest neighbors (KNN), and artificial neural networks.
• Application of algorithms to a thyroid disease dataset, including manipulated (sampled and unsampled) versions.
• Evaluation of classification performance based on accuracy and specificity metrics.

Main Results:

• The random forest algorithm demonstrated the highest performance after dataset manipulation.
• Achieved 94.8% accuracy and 91% specificity using the random forest algorithm.
• Comparative analysis highlighted the superiority of certain machine learning models in thyroid disease classification.

Conclusions:

• Machine learning, particularly the random forest algorithm, offers a highly accurate approach for classifying thyroid diseases from ultrasound data.
• Dataset manipulation can significantly enhance the predictive performance of machine learning models for thyroid conditions.
• Automated classification systems hold potential to reduce radiologist workload and improve diagnostic accuracy.