Contrast dose reduction for medical imaging using deep learning
A method for diagnostic imaging with reduced contrast agent dose uses a deep learning network (DLN) [ 114 ] that has been trained using zero-contrast [ 100 ] and low-contrast [ 102 ] images as input to the DLN and full-contrast images [ 104 ] as reference ground truth images. Prior to training, the images are pre-processed [ 106, 110, 118 ] to co-register and normalize them. The trained DLN [ 114 ] is then used to predict a synthesized full-dose contrast agent image [ 116 ] from acquired zero-dose and low-dose images.
1. A method for improving quality of a medical image of a subject, the method comprising:
a) acquiring a first image with a first image acquisition sequence, wherein the first image is acquired with zero contrast agent dose administered to the subject;
b) acquiring a second image with a second image acquisition sequence distinct from the first image acquisition sequence, wherein the first image and the second image are acquired using a common imaging modality, wherein the second image is acquired with zero contrast agent dose administered to the subject;
c) processing the first image and the second image to adjust for acquisition and scaling differences; and
d) applying the first image and the second image as input to a deep learning network (DLN) to generate as output of the DLN an image of the subject with an enhanced quality.
2. The method of claim 1 wherein the DLN is trained using a training dataset including zero-contrast agent dose images acquired with different imaging sequences and full-contrast agent dose images acquired with a full contrast agent dose administered as reference ground-truth images.
3. The method of claim 1 wherein the DLN is trained using a training dataset including zero-contrast agent dose images acquired with zero contrast agent dose, low-contrast agent dose images acquired with low contrast agent dose and full-contrast agent dose images acquired with a full contrast agent dose administered as reference ground-truth images.
4. The method of claim 1 wherein the zero-contrast agent dose images are acquired with different imaging sequences.
5. The method of claim 1 wherein the low contrast agent dose is less than 10% of a full contrast agent dose.
6. The method of claim 1 wherein the common imaging modality is selected from the group consisting of angiography, fluoroscopy, computed tomography (CT), ultrasound, and magnetic resonance imaging.
7. The method of claim 1 wherein the common imaging modality is magnetic resonance imaging, and a full contrast agent dose is at most 0.1 mmol/kg Gadolinium MRI contrast.
8. The method of claim 1 wherein the DLN is an encoder-decoder convolutional neural network (CNN) including bypass concatenate connections and residual connections.
9. The method of claim 1 wherein the first image acquisition sequence is selected from a T1w, T2w, FLAIR, or DWI sequence.
10. The method of claim 1 wherein the second image acquisition sequence is selected from a T1w, T2w, FLAIR, or DWI sequence.
11. The method of claim 1 wherein processing the first image and the second image to co-register and normalize the images comprises normalizing the intensity within each contrast via histogram matching on median.
12. The method of claim 1 wherein processing the first image and the second image comprises including a bias field correction to correct the bias field distortion via a N4ITK algorithm.
13. The method of claim 1 wherein processing the first image and the second image comprises using a trained classifier (VGG16) to filter out abnormal slices.