Thyroid Diagnosis from SPECT Images Using Convolutional Neural Network with Optimization.

Area of Science:

• Endocrinology
• Medical Imaging
• Artificial Intelligence

Background:

• Thyroid disease is a prevalent endocrine disorder, with SPECT imaging crucial for clinical diagnosis.
• Limited research exists on applying machine learning to computer-aided diagnosis of thyroid diseases using SPECT images.

Purpose of the Study:

• To develop an optimized convolutional neural network (CNN) for computer-aided diagnosis of thyroid diseases using SPECT images.
• To evaluate the CNN's efficacy in distinguishing between Graves' disease, Hashimoto disease, and subacute thyroiditis.

Main Methods:

• A modified DenseNet architecture was employed, incorporating trainable weight parameters in skip connections.
• The learning rate was optimized using the flower pollination algorithm for enhanced network training.
• The model was trained and validated on SPECT images for three distinct thyroid disease categories.

Main Results:

• The proposed CNN method demonstrated high efficiency in diagnosing thyroid diseases from SPECT images.
• The optimized CNN achieved superior performance compared to existing CNN-based diagnostic methods.
• The modified architecture and training strategy significantly improved diagnostic accuracy.

Conclusions:

• The developed machine learning approach offers an efficient and effective tool for thyroid disease diagnosis using SPECT imaging.
• This study highlights the potential of optimized CNNs in advancing computer-aided diagnosis in endocrinology.