PET Parameter Image Enhancement via Iterative Deep Learning

The patent replaces traditional mechanical filtering algorithms with a deep learning-based image enhancement model. The model uses convolutional neural networks to learn the mapping between low-quality and high-quality images, substituting the mechanical filtering process with an intelligent system that can preserve spatial resolution while reducing noise through learned patterns rather than fixed mathematical operations.

The patent changes the fundamental parameters of the image enhancement process by using a trained neural network model with optimized weights and biases. Instead of applying fixed filtering parameters, the system uses learned parameters from training data to dynamically adjust the enhancement process, allowing simultaneous noise reduction and detail preservation through adaptive parameter selection.

The patent uses synthetic data generation to create artificial high-quality PET parameter images that serve as training labels. Instead of requiring actual clinical high-quality scans, the system generates synthetic training data through simulation and computational methods, copying the essential characteristics needed for training without the time and resource costs of acquiring real high-quality medical images.

The patent replaces expensive, time-consuming real high-quality medical images with computationally generated synthetic images. These synthetic training labels are inexpensive to produce and can be generated rapidly without requiring actual patient scanning time or additional tracer doses, effectively substituting valuable clinical resources with affordable computational alternatives.

The patent generates synthetic training data that replicates the essential features of high-quality PET images without requiring actual high-dose tracer injections. The synthetic data copying approach creates virtual training examples that preserve the statistical and structural properties needed for model training while eliminating the need for additional radioactive tracer administration to patients.

The patent substitutes expensive radioactive tracer materials with computationally generated synthetic data. The synthetic training labels serve as disposable, inexpensive alternatives to real high-dose images, eliminating the need to consume additional tracer substances while providing sufficient training material for the neural network model.