Aim/PurposeThe purpose of this research is building a deep learning framework for thyroid medical image analysis by using CNNs.

BackgroundThyroid disease diagnosis depends critically on thyroid medical image analysis. Though their design and settings can be optimized, Convolutional Neural Networks (CNNs) have showed promise in automating this process. Powerful metaheuristic technique Particle Swarm Optimization (PSO) can efficiently optimize CNNs for improved performance.

MethodologyWe provide in this work a deep learning method for thyroid medical image interpretation and prediction by merging CNNs with PSO. In order to extract pertinent features, we first preprocess the medical images. We next construct a CNN architecture based on the image properties. Using PSO, the CNN architecture and its hyperparameters—filter sizes, layer count, and learning rates—are optimized. To learn discriminative features and patterns, the optimized CNN is trained using a sizable dataset of thyroid images.

ContributionOur method uses the synergistic potential of deep learning and metaheuristic optimization to advance the field of medical image analysis. Improved thyroid image analysis task performance results from our efficient search of the large space of CNN architectures and hyperparameters by integrating PSO.

FindingsOur proposed method is shown to be beneficial by experimental data. Higher accuracy, sensitivity, and specificity were obtained by the optimized CNN in thyroid disease detection than by conventional CNN architectures.

Recommendation for ResearchersFurthermore, in terms of computing efficiency and accuracy of prediction, our method beat those of previous methods. Furthermore improved interpretability of the learnt characteristics gave important new perspectives on the underlying patterns in thyroid images.

Future ResearchIn future research, this work can be enhanced using several deep learning algorithms for achieving better accuracy and performance.