Specific Binding Ratio Estimation of [¹²³I]-FP-CIT SPECT Using Frontal Projection Image and Machine Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nuclear Medicine

Background:

• Single-photon emission-computed tomography (SPECT) is crucial for diagnosing neurodegenerative diseases.
• Estimating the specific binding ratio (SBR) is vital for quantitative analysis in SPECT imaging.
• Current SBR estimation methods can be time-consuming and require complex preprocessing.

Purpose of the Study:

• To develop and validate a novel convolutional neural network (CNN) for estimating SBR.
• To assess the feasibility of using only frontal projection images for SBR estimation.
• To evaluate the accuracy and efficiency of the proposed CNN method.

Main Methods:

• Development of two CNN models, LeNet and AlexNet, trained on five distinct datasets derived from [123I]ioflupane frontal projection images.
• Implementation of data augmentation techniques, including left-right reversal, to enhance training data.
• Quantitative assessment of SBR estimation accuracy using mean absolute error, root mean squared error, correlation coefficient, and slope.

Main Results:

• The proposed CNN method demonstrated high accuracy in estimating SBR from frontal projection images alone.
• The best correlation coefficient achieved between SPECT-derived SBR and estimated SBR was 0.87.
• CNN models trained on cropped and augmented data (40FOV_DAhalf) showed superior performance compared to the unprocessed 128FOV dataset.

Conclusions:

• The developed CNN method offers a feasible and accurate approach for SBR estimation in clinical settings.
• Utilizing only frontal projection images significantly reduces acquisition time and complexity.
• This AI-driven method holds promise for efficient and reliable quantitative analysis in [123I]ioflupane SPECT imaging.