Neural network evolution using expedited genetic algorithm for medical image denoising
View Patent ↗Various embodiments for image denoising using a convolutional neural network (CCN) are described. A system may include at least one computing device and program instructions stored in memory and executable in the at least one computing device that, when executed, direct the at least one computing device to implement a genetic algorithm (GA) routine that identifies and optimizes a plurality of hyperparameters for use in denoising an image using the convolutional neural network. An image may be denoised using the convolutional neural network, where the image is denoised using the hyperparameters identified and optimized in the genetic algorithm routine.
1. A system for image denoising using a convolutional neural network (CCN), comprising:
at least one computing device comprising at least one hardware processor; and
program instructions stored in memory and executable in the at least one computing device that, when executed, direct the at least one computing device to:
perform a genetic algorithm (GA) routine that identifies and optimizes a plurality of hyperparameters for use in denoising an image using the convolutional neural network (CNN), wherein at least one of the plurality of hyperparameters identified and optimized comprises a number of layers, a number of neurons in each layer, and a type of activation function to be employed in the convolutional neural network (CNN); and
denoise an image using the convolutional neural network (CCN), wherein the image is denoised by the convolutional neural network (CNN) using the plurality of hyperparameters identified and optimized in the genetic algorithm (GA) routine.
2. The system of claim 1 wherein the genetic algorithm (GA) routine utilizes a fitness score determined for each set of the plurality of hyperparameters.
3. The system of claim 2 , wherein the fitness score is a restored image quality metric determined using a fitness function.
4. The system of claim 1 , wherein the genetic algorithm routine is configured to perform at least one of: a selection operation, a crossover operation, and a mutation operation.
5. The system of claim 1 , wherein the at least one computing device is further directed to split a set of all possible hyperparameters into (i) a fine-gene set and (ii) a complementary-gene set, where the fine-gene set comprises at least one other hyperparameter selected from a prior convolutional neural network.
6. The system of claim 5 , wherein:
a first population of the genetic algorithm routine is initialized based on the fine-gene set; and
a mutation operation is performed using the complementary-gene set.
7. The system of claim 1 , wherein the at least one computing device is further directed to implement a transfer learning routine by performing the genetic algorithm routine on a first dataset and applying the hyperparameters as identified from a result of the first dataset to a second dataset, wherein the second dataset is larger than the first dataset.
8. The system of claim 1 , wherein the image is a medical image, the medical image comprising a magnetic resonance imaging (MRI) image, a computed tomography (CT) image, or an X-ray image.
9. The system of claim 1 , wherein:
the at least one of the plurality of hyperparameters identified and selected is a first subset of the plurality of hyperparameters; and
a second subset of the hyperparameters comprises at least one of: a number of filters, a type of nonlinear activation function, and a choice of optimizer function to be employed in the convolutional neural network (CNN).
10. The system of claim 1 , wherein the at least one computing device is further directed to optimize an evolutionary process of the genetic algorithm (GA) using at least one of an experience-based greedy exploration routine and a transfer learning routine.
11. A computer-implemented method, comprising:
performing, by at least one computing device that comprises at least one hardware processor, a genetic algorithm (GA) routine that identifies and optimizes a plurality of hyperparameters for use in denoising an image using the convolutional neural network (CNN), wherein at least one of the hyperparameters identified and optimized comprises a number of layers, a number of neurons in each layer, and a type of activation function to be employed in the convolutional neural network (CNN); and
denoising, by the at least one computing device, an image using the convolutional neural network (CCN), wherein the image is denoised by the convolutional neural network (CNN) using the hyperparameters identified and optimized in the genetic algorithm (GA) routine.
12. The computer-implemented method of claim 11 , further-comprising utilizing, by the at least one computing device, a fitness score determined for each set of the plurality of hyperparameters in the genetic algorithm (GA) routine.
13. The computer-implemented method of claim 12 , further comprising:
determining, by the at least one computing device, a restored image quality metric using a fitness function; and
selecting, by the at least one computing device, the restored image quality metric as the fitness score.
14. The computer-implemented method of claim 11 , further comprising performing, by the at least one computing device in the genetic algorithm (GA) routine, at least one of: a selection operation, a crossover operation, and a mutation operation.
15. The computer-implemented method of claim 11 , further comprising splitting, by the at least one computing device, a set of all possible hyperparameters into (i) a fine-gene set and (ii) a complementary-gene set, where the fine-gene set comprises at least one other hyperparameter selected from a prior convolutional neural network.
16. The computer-implemented method of claim 15 , wherein:
a first population of the genetic algorithm routine is initialized based on the fine-gene set; and
a mutation operation is performed using the complementary-gene set.
17. The computer-implemented method of claim 11 , further comprising:
implementing, by the at least one computing device, a transfer learning routine by performing the genetic algorithm (GA) routine on a first dataset and applying the hyperparameters as identified from a result of the first dataset to a second dataset, wherein the second dataset is larger than the first dataset.
18. The computer-implemented method of claim 11 , wherein the image is a medical image, the medical image comprising a magnetic resonance imaging (MRI) image, a computed tomography (CT) image, or an X-ray image.
19. The computer-implemented method of claim 11 , wherein:
the at least one of the plurality of hyperparameters identified and selected is a first subset of the plurality of hyperparameters; and
a second subset of the hyperparameters comprises at least one of: a number of filters, a type of nonlinear activation function, and a choice of optimizer function to be employed in the convolutional neural network (CNN).
20. The computer-implemented method of claim 11 , further comprising optimizing, by the at least one computing device, an evolutionary process of the genetic algorithm (GA) using at least one of an experience-based greedy exploration routine and a transfer learning routine.