Authors:
S. Venkatasubramanian
Addresses:
Department of Computer Science and Business Systems, Saranathan College of Engineering, Trichy, Tamil Nadu, India.
Abstract:
Thyroid disease (TD) patients benefit greatly from early diagnosis and treatment. Ultrasounds and blood tests are two important diagnostic tools for evaluating thyroid function abnormalities. The seriousness of thyroid issues necessitates the use of reliable diagnostic methods. This paper introduces a new approach to TD classification by utilising cutting-edge feature extraction, classification, and picture preprocessing methods. Thyroid image noise is efficiently reduced by the Fast Non-Local Means Algorithm (FNLM), providing a clean input for further analysis. Histogram Features is a strategy for creating a comprehensive representation for disease classification that extracts discriminative features from pre-processed images. To enhance the model's ability to identify subtle patterns that might indicate thyroid issues, propose using a Triple Attention Guided Residual Dense (TriAG-RDBNet) and a BiLSTM-based connection model. The CTOA (Chaotic-Based Tumbleweed Optimisation) Algorithm is used to fine-tune the model's hyperparameters for optimal execution. Dynamic parameter optimisation helps the model achieve the highest possible classification accuracy. Extensive testing on a publicly available dataset demonstrates the efficiency of the proposed method, achieving an astounding 99.22% accuracy. Our integrated approach outperforms state-of-the-art models in comparison tests, suggesting it might be a helpful tool for TD classification.
Keywords: Bidirectional Long Short-Term Memory; Tumbleweed Optimisation; Diagnostic Tools; Thyroid Disease (TD); Histogram Features; Diagnosis and Treatment; Classification Accuracy.
Received: 21/08/2024, Revised: 08/12/2024, Accepted: 24/01/2025, Published: 05/06/2025
DOI: 10.64091/ATIHL.2025.000167
AVE Trends in Intelligent Health Letters, 2025 Vol. 2 No. 2 , Pages: 72-84