A hybrid thyroid tumor type classification system using feature fusion, multilayer perceptron and bonobo optimization

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Thyroid cancer, though historically rare, is increasingly prevalent.
• Early and accurate diagnosis is crucial for effective treatment.

Purpose of the Study:

• To develop a novel hybrid artificial intelligence system for classifying four types of thyroid tumors.
• To enhance classification performance using advanced image augmentation and feature fusion techniques.

Main Methods:

• Ultrasound images were augmented using data warping techniques.
• Image preprocessing involved bilateral filtering and dynamic histogram equalization.
• Segmentation was performed using SegNet, with feature extraction via CapsuleNet and EfficientNetB2, followed by fusion.
• Classification was achieved using a Multilayer Perceptron Classifier optimized by Bonobo optimizer.

Main Results:

• The proposed hybrid system demonstrated high classification performance.
• Key performance metrics included accuracy, sensitivity, specificity, F1-score, and Matthew's correlation coefficient.
• The system outperformed existing classifiers such as CANFES, Spatial Fuzzy C means, Deep Belief Networks, Thynet, Generative Adversarial Network, and Long Short-Term Memory.

Conclusions:

• The proposed Multilayer Perceptron-based thyroid tumor classification system is efficient and effective.
• This AI-driven approach offers a promising advancement in thyroid tumor diagnosis.