Differentiation of the Follicular Neoplasm on the Gray-Scale US by Image Selection Subsampling along with the...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Thyroid nodules require accurate differentiation between benign adenoma and malignant carcinoma.
• Current diagnostic methods can be limited, necessitating advanced analytical approaches.

Purpose of the Study:

• To develop and evaluate a deep learning model for differentiating thyroid follicular adenoma from carcinoma using ultrasonography (US) images.
• To investigate the efficacy of a novel normalization technique in enhancing diagnostic accuracy.

Main Methods:

• Utilized 8-bit bitmap ultrasonography images of thyroid nodules.
• Applied a convolutional neural network (CNN) with a majority decision aggregation.
• Implemented a scalable, parameterized normalization treatment for image data.
• Trained the CNN model on a small dataset of 39 benign adenomas and 39 carcinomas.

Main Results:

• Achieved an overall differentiation accuracy of 89.51% on the test dataset.
• Obtained 93.19% accuracy for benign adenoma and 71.05% for carcinoma.
• Demonstrated high performance even with extremely small training datasets.
• Presented numerical results including area under the receiver operating characteristic (AUROC) curve.

Conclusions:

• Deep learning, specifically CNNs, can effectively differentiate thyroid follicular adenoma from carcinoma using US images.
• The proposed normalization method shows promise in improving diagnostic accuracy.
• The findings suggest the potential for AI-driven tools to aid in thyroid nodule diagnosis.