3-D Convolutional Autoencoder for Low-Dose CT via 2-D Transfer Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low-dose computed tomography (CT) images often suffer from noise and artifacts due to reduced radiation doses, which can impair diagnostic performance, and existing methods struggle to effectively reduce these issues while maintaining image quality.
Innovation Solution
A 3-D convolutional autoencoder is trained using transfer learning from a 2-D network, incorporating a Wasserstein generative adversarial network with gradient penalty and perceptual loss functions to correct low-dose CT images and approximate normal-dose CT images by reducing noise and artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If radiation dose is reduced to minimize cancer risk, then patient safety is improved, but image quality deteriorates due to increased noise and artifacts
Solution Approach 1:
A neural network model serves as an intermediary between the low-dose CT images and the desired high-quality normal-dose images. The network learns the mapping relationship from noisy low-dose images to clean normal-dose images through training on paired data, effectively mediating the quality degradation caused by radiation dose reduction.
Solution Approach 2:
The system creates a digital copy of the low-dose CT image and processes it through the neural network to generate a synthetic normal-dose image. This copying approach allows the original low-dose image to remain unchanged while producing an enhanced version that mimics the quality of higher-dose scans.
2Measurement precision
If 3-D data is used for training to capture spatial relationships, then image correction accuracy is improved, but computational cost and training time increase significantly
Solution Approach 1:
The 3-D image correction problem is segmented into multiple 2-D slice processing tasks. Instead of training a single 3-D convolutional network that processes entire volumetric datasets, the system divides the volume into individual axial slices, processes each slice independently through 2-D convolutional layers, and then combines the results. This segmentation dramatically reduces the computational burden while preserving spatial relationships through the sequential processing of adjacent slices.
Solution Approach 2:
The approach transforms the problem from a 3-D volumetric processing domain to a 2-D slice processing domain. By reducing the input dimensionality from volumetric 3-D data to planar 2-D slices, the system lowers the computational complexity of convolutional operations while maintaining the essential spatial information needed for accurate image correction.
Data Source
AI summary
A 3-D convolutional autoencoder for low-dose CT via transfer learning from a 2-D trained network is described, A machine learning method for low dose computed tomography (LDCT) image correction is provided. The method includes training, by a training circuitry, a neural network (NN) based, at least in part, on two-dimensional (2-D) training data. The 2-D training data includes a plurality of 2-D training image pairs. Each 2-D image pair includes one training input image and one corresponding target output image. The training includes adjusting at least one of a plurality of 2-D weights based, at least in part, on an objective function. The method further includes refining, by the training circuitry, the NN based, at least in part, on three-dimensional (3-D) training data. The 3-D training data includes a plurality of 3-D training image pairs. Each 3-D training image pair includes a plurality of adjacent 2-D training input images and at least one corresponding target output image. The refining includes adjusting at least one of a plurality of 3-D weights based, at least in part, on the plurality of 2-D weights and based, at least in part, on the objective function. The plurality of 2-D weights includes the at least one adjusted 2-D weight.