An Improved k-Nearest Neighbor Algorithm for Recognition and Classification of Thyroid Nodules.

Area of Science:

• Medical imaging analysis
• Machine learning in diagnostics

Background:

• Thyroid nodule classification is crucial for diagnosis.
• Imbalanced datasets pose challenges for traditional classification methods.
• High classification error rates impact diagnostic accuracy.

Purpose of the Study:

• To develop an automatic thyroid nodule recognition and classification system.
• To address the issue of high classification error rates in imbalanced sample scenarios.
• To improve the accuracy and reliability of thyroid nodule classification.

Main Methods:

• An improved k-nearest neighbor (KNN) algorithm was developed.
• The enhanced KNN algorithm incorporates class label counts and weights.
• Minkowski distance measure was utilized instead of Euclidean distance.

Main Results:

• The study analyzed 508 thyroid nodule ultrasound images (415 benign, 93 malignant).
• The improved KNN achieved an accuracy of 0.8725, precision of 0.8673, recall of 1, and F1-score of 0.9290.
• The influence of distance weights, k value, and distance measures on classification was evaluated.

Conclusions:

• The proposed improved KNN method demonstrates superior performance.
• The method significantly outperforms traditional KNN and other classical machine learning approaches.
• This advancement offers a more reliable tool for automatic thyroid nodule classification.