Denoising medical images using machine learning

Abstract

A method for denoising medical images. The medical image is rectified or normalised using a noise map that defines estimates of one or more statistical parameters for each pixel of the medical image. The rectified medical image is then processed using machine learning methods to produce a denoised medical image.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging, and more particularly to noise removal of medical images.

Background Art

[0002] Medical imaging is gaining increasing interest among medical professionals to assist (non-invasively) in the evaluation and / or diagnosis of the condition of a subject or patient under examination. Various forms of medical imaging techniques and modalities are known in the art, employing invasive or non-invasive imaging techniques. Examples include computed tomography (CT) or X-ray imaging, magnetic resonance (MR) imaging, (intravascular) ultrasound imaging, positron emission tomography PET imaging, optical coherence tomography, transesophageal echocardiogram, and the like.

[0003] Ongoing concerns regarding medical images are noise and artifacts. When evaluating the condition of a subject, noise and artifacts can interfere with the identification of potentially important features of the subject (e.g., by obscuring them or making identification difficult) and can be mistaken for diagnostically relevant features. Thus, there is a continuing desire to reduce the amount of arch noise in medical images.

[0004] One recently developed approach for accurately removing noise from medical images is to use a properly trained machine learning method to perform noise removal. However, there are a number of parameters that can be changed during the medical imaging process. For example, in a CT scan, various forms of reconstruction filters can be used to generate CT medical images. Machine learning methods tend to overfit to the training data and, therefore, often cannot generalize to a wide range of real-world parameter settings (which are not necessarily sampled during training).

Summary of the Invention

Problems to be Solved by the Invention

[0005] Therefore, improved techniques are needed for noise reduction in medical images. [Means for solving the problem]

[0006] The present invention is defined by the claims.

[0007] According to one embodiment of the present invention, a computer-based method is provided for denoising medical images and generating denoised medical images.

[0008] The computer implementation method includes the steps of: acquiring a medical image formed from multiple pixels; acquiring a noise map containing estimated measurements of statistical parameters for each pixel of the medical image; modifying the medical image using the noise map to generate a modified medical image; and processing the modified medical image using a machine learning method to generate a denoised medical image.

[0009] The proposed invention proposes a method for normalizing / uniforming the global (i.e., image width) noise level of medical images based on statistical parameters of the noise in the medical images. The normalized / uniformed medical images are then processed using machine learning methods to perform, for example, further noise reduction.

[0010] This ensures that machine learning methods consistently process medical images that have already been normalized or homogenized based on local / global conditions. This means that the images provided to the machine learning method are at a consistent noise level and / or homogenized (e.g., to decorrelate the noise). Thus, the machine learning method is trained on normalized / homogenized medical images, thus avoiding the problem of overfitting.

[0011] Therefore, the proposed method provides a way to de-noise, improve, and more accurate medical images.

[0012] The step of correcting a medical image may include dividing the medical image by a noise map. This division may be on a pixel-by-pixel or point-by-point basis.

[0013] In some embodiments, the step of processing a corrected medical image includes processing the corrected medical image using a machine learning method to generate a predictive noise image, wherein the predictive noise image represents the amount of noise predicted at each pixel of the corrected medical image; multiplying the corrected medical image by a noise map to generate a calibrated predictive noise image; and subtracting the calibrated predictive noise image or a scaled version of the calibrated predictive noise image from the medical image to generate a denoised medical image.

[0014] In some embodiments, the step of processing a corrected medical image includes inputting the corrected medical image into a machine learning method and receiving a denoised medical image as output from the machine learning method. The denoised medical image output by the machine learning method may not be calibrated (for example, if the medical image has been corrected by dividing it by a noise map). Therefore, the denoised medical image output by the machine learning method may be recalibrated using a noise map. Specifically, the denoised medical image can be recalibrated by applying the opposite (or reverse) of the correction made to the medical image to generate the corrected medical image to the denoised medical image. This may include, for example, multiplying the denoised medical image by a noise map (for example, on a pixel-by-pixel basis) to generate a recalibrated denoised medical image.

[0015] The machine learning method may include or could be a neural network. This provides an accurate and reliable method for processing corrected medical images to perform the denoising process.

[0016] The machine learning method is preferably a residual (output) machine learning method, such as a residual neural network. In the context of this disclosure, the residual machine learning method is a method for processing a corrected medical image to provide a predicted noise image that shows the amount of noise predicted for each pixel of the corrected medical image.

[0017] The use of residual machine learning methods is advantageous because it makes the output of the neural network more reliable. Specifically, the predicted noisy image can be assumed to be within the range of the output for which the machine learning method was trained (since noise only has a limited range of estimates). Non-residual or direct machine learning methods (e.g., directly inferring denoised medical images) are more likely to result in predicted denoised images that are outside the range for which the machine learning method was trained.

[0018] A noise map can provide an estimate of the standard deviation or variance of noise at each pixel in a medical image.

[0019] In another embodiment, the noise map provides an estimated correlation between the noise of each pixel in a medical image and the noise of one or more adjacent pixels.

[0020] In another embodiment, the medical image is one of several medical images generated by a multi-channel imaging process that represent the same scene, and the noise map provides an estimated value of the covariance or correlation between the noise of each pixel in the medical image and the noise of the corresponding pixel in another medical image among the several medical images.

[0021] This technique allows for more effective reduction of inter-channel crosstalk in multi-channel imaging processes, or enables consideration of it when denoising medical images (from multi-channel imaging processes). It will be understood that each medical image from each channel, i.e., each of the multiple images, can be processed separately using the method described herein.

[0022] In some embodiments, the step of acquiring a medical image includes the steps of acquiring a first medical image, processing the first medical image using a frequency filter to acquire a filtered medical image having values ​​within a predetermined frequency range, and setting the first filtered medical image as a medical image.

[0023] In at least one embodiment, the method further includes the steps of processing a first medical image to obtain a second filtered medical image having values ​​in a second different predetermined frequency range (e.g., not including any of the first frequency ranges), and combining the second filtered medical image and the denoised medical image to generate a denoised first medical image.

[0024] In a preferred embodiment, the medical image is a medical image reconstructed from raw data using a first reconstruction algorithm, and the machine learning method is a machine learning method trained using a training dataset containing one or more training images reconstructed from raw data using a second different reconstruction algorithm. In such embodiments, it is recognized that different reconstruction filters result in different noise characteristics. By homogenizing the medical images processed using the machine learning method, the machine learning method trained using medical images generated with different reconstruction filters can still be used with high reliability.

[0025] In some embodiments, the machine learning method is preferably trained using a training dataset that includes one or more training images, each modified with a corresponding noise map, in the same manner as the medical images processed using the machine learning method. This approach improves the suitability and reliability of the machine learning method.

[0026] The medical image can be a computed tomography medical image. The proposed method is recognized as being particularly useful for medical images that can be generated using different reconstruction filters, and thus is particularly useful in the use with computed tomography images that can use various reconstruction filters.

[0027] A computer program product including computer program code means is also proposed. When the computer program code means is executed on a computing device having a processing system, it causes the processing system to perform all steps of any of the methods described herein.

[0028] Also proposed is a processing system configured to remove noise from a medical image and generate a noise-removed medical image. The processing system acquires a medical image formed from a plurality of pixels, acquires a noise map including estimated measurements of statistical parameters of each pixel of the medical image, modifies the medical image using the noise map to generate a modified medical image, and is configured to process the modified medical image using a machine learning method to generate a noise-removed medical image.

[0029] A system is also provided that includes the above-described processing system and a medical imaging system configured to generate a medical image and provide the generated medical image to the processing system.

[0030] These and other aspects of the invention will be made apparent from the embodiments described below and will be described with reference to the embodiments.

[0031] To better understand the present invention and to more clearly show how it is implemented, reference is made, by way of example only, to the accompanying drawings.

Brief Description of the Drawings

[0032] [Figure 1] It is a flowchart for explaining the method according to the embodiment. [Figure 2]This is a flowchart explaining the process used in the method. [Figure 3] This is a flowchart illustrating the method according to the embodiment. [Figure 4] This section demonstrates the operation of the proposed method. [Figure 5] The processing system according to the embodiment is shown. [Figure 6] The system according to the embodiment is shown. [Modes for carrying out the invention]

[0033] The present invention will be described with reference to the figures.

[0034] The detailed descriptions and specific examples illustrate exemplary embodiments of the apparatus, system, and method, but should be understood to be for illustrative purposes only and not intended to limit the scope of the invention. These and other features, aspects, and advantages of the apparatus, system, and method of the present invention will be better understood from the following description, the appended claims, and the appended drawings. It should be understood that the figures are schematic diagrams only and are not drawn to scale. It should also be understood that the same reference numerals are used throughout the figures to indicate the same or similar parts.

[0035] This invention provides a method for denoising medical images. The medical image is modified or normalized using a noise map that defines estimates of one or more statistical parameters for each pixel of the medical image. The modified medical image is then processed using a machine learning method to generate a denoised medical image.

[0036] The embodiment is based on the recognition that the reconstruction filters or algorithms used to generate medical images result in noise in differently reconstructed medical images with different statistical properties. This reduces the efficiency of denoising machine learning methods because they may not be trained on images generated using the same reconstruction filters. By correcting medical images using estimates of statistical parameters, each medical image can effectively have a normalized level of noise. This greatly improves the consistency, and therefore the accuracy, of the denoising machine learning method.

[0037] The proposed concept can be used in any medical imaging system that can be used in a wide range of clinical settings.

[0038] In the context of this disclosure, medical images refer to images obtained using medical imaging modalities such as X-ray images, CT (computed tomography) images, PET (positron emission tomography) images, MR (magnetic resonance imaging) images, or ultrasound images. Other forms of medical images will be apparent to those skilled in the art.

[0039] Figure 1 is a flowchart illustrating a method 100 for noise reduction of a medical image 105 according to an embodiment.

[0040] Method 100 can be performed, for example, on a single image. In another example, Method 100 can be performed on one or more (e.g., each) of multiple images. Since multiple medical images are generated by a multi-channel imaging process, each image may represent a different channel and represent the same scene. In other words, each of the multiple images may represent the same anatomical region.

[0041] Method 100 includes step 110 of acquiring a medical image to denoise. The medical image is formed of multiple pixels and may be two-dimensional (2D) or three-dimensional (3D). The pixels in the 3D image may be labeled as voxels.

[0042] Step 110 may itself include generating a medical image using, for example, a appropriately configured medical imaging device. In other embodiments, step 110 includes retrieving the already generated medical image from, for example, the medical imaging device and / or memory or storage unit.

[0043] Method 100 further includes step 120 of obtaining a noise map 107 of the medical image. The noise map contains estimated values ​​of the statistical parameters of noise for each pixel of the medical image. In other words, the noise map defines, for each pixel of the medical image, an estimate of the statistical parameters of noise for that pixel. It should be emphasized that the noise map provides estimated statistical values ​​of noise, not the specific intensity at each pixel of the medical image.

[0044] A noise map may contain dedicated estimated values ​​for each pixel of a medical image (for example, one made up of the same number of pixels as the medical image). In another embodiment, a noise map may contain estimated values ​​for groups of two or more pixels, so that a single estimated value represents the statistical parameter of noise for multiple pixels. Thus, a single estimated value may represent the statistical parameter of noise for each region of the medical image.

[0045] In some embodiments, the noise map may include estimates of the noise variance and / or noise standard deviation for each pixel of a medical image. Techniques for generating such noise maps are established in the art. Several examples are disclosed in the U.S. Patent No. 10,984,564(B2). Other examples include U.S. Patent No. 9,1591,22(B2), filed November 12, 2012, titled "Image domain de-noising," U.S. Patent Application Publication No. 2016 / 0140725(A1), filed June 26, 2014, titled "Methods of utilizing image noise information," and U.S. Patent No. 8,938,110(B2), filed October 29, 2015, titled "Enhanced image data / dose reduction."

[0046] In another example, a noise map may include estimated values ​​of the correlation between the noise of a given pixel / region and the noise of adjacent pixels / regions (for each pixel / region). A coarse estimate of the noise in an image or image region can be generated using any coarse denoising method known in the art. From such noise estimates, local noise correlations can be directly estimated by performing a standard correlation analysis.

[0047] As yet another example, a noise map may include a noise probability density function (noise PDF) for each pixel or region of a medical image. The noise PDF can be used, for example, to determine or predict the noise variance and / or standard deviation of a pixel / region.

[0048] As yet another example, a noise map may include the covariance or correlation between the noise of one pixel and the noise of a corresponding pixel in another medical image. The medical image and other medical images may represent the same scene and form part of multiple medical images generated, for example, by a multi-channel image processing.

[0049] Step 120 may itself include the step of generating a noise map using, for example, one of the methods described above. Alternatively, step 120 may include obtaining an already generated noise map of a medical image from, for example, memory or a storage unit.

[0050] Furthermore, in method 100, step 130 is performed to modify the medical image using a noise map in order to generate a corrected medical image.

[0051] Step 130 may include, for example, normalizing the medical image using estimated noise statistics. This aims to effectively normalize or normalize these statistics across the entire medical image.

[0052] For example, step 130 includes dividing the medical image by a noise map. Specifically, the value of each pixel in the medical image is divided by an estimated value of the noise statistical parameter for that pixel, provided by the noise map. In this way, point-level or pixel-level division of the medical image by the estimated noise statistical parameter can be performed.

[0053] As another example, step 130 includes processing the noise map to obtain statistical information about the noise map and / or the medical image. This statistical information can then be used to modify the medical image.

[0054] For example, a noise map can be processed to determine the mean estimated deviation of the measured values ​​of a statistical parameter (e.g., for the entire image or for different sections of the image). Then, the medical image can be divided by the mean estimated deviation. For example, it can be divided by each section for the mean representing the mean of a particular section, or by the entire image if the mean represents the mean of the entire image.

[0055] As yet another example, noise maps can be processed to determine or predict the shape of the noise in the frequency domain. This shape can then be used to perform frequency domain filtering on medical images to normalize the frequency of the noise in those images, for example.

[0056] As another example, step 130 includes performing a decorrelation process on the noise in the medical image, i.e., a process to decorrelate the noise in the medical image. Thus, step 130 includes (spatially) decorrelation of the noise in the medical image in order to generate a corrected medical image.

[0057] This can be done using a noise map that shows measured correlations between different pixels and / or regions. Alternatively, this can be done by processing the image using one or more means of approximate deconvolution or transformation of the PDF by passing the values ​​through some nonlinear function, using a noise map that provides a noise PDF for each pixel and / or region.

[0058] Those skilled in the art will readily conceive of various other techniques for correcting the medical image in step 130. More generally, step 130 is the step of correcting the medical image so that the noise response of the entire corrected medical image is more uniform than that of the entire (original) medical image.

[0059] The output of step 130 is the corrected medical image 135.

[0060] Next, method 100 performs a process 140 in which the corrected medical image is processed using a machine learning method in order to generate a denoised medical image.

[0061] Figure 1 shows one embodiment for performing process 140.

[0062] In this example, process 140 includes directly predicting or inferring a denoised medical image 145 from a corrected medical image 135. Thus, the machine learning method takes a corrected medical image as input and provides a denoised medical image as output. Thus, the machine learning method can be trained to generate a clean or denoised image from a (noisy) medical image.

[0063] In some embodiments, the machine learning method outputs only uncalibrated denoised medical images. The uncalibrated denoised medical images are (re)calibrated using a noise map. Specifically, the uncalibrated denoised medical images are modified to generate denoised medical images by reversing the procedure performed in step 130 using the uncalibrated denoised medical images and the noise map.

[0064] For example, if step 130 includes performing pixel-level division of a medical image using a noise map, then pixel-level multiplication using the noise map is performed on an uncalibrated, denoised medical image.

[0065] (Re)calibration may not be necessary depending on the circumstances and embodiment, such as when step 130 involves decorrelation of noise in the medical image.

[0066] Figure 2 shows another embodiment for performing process 140, which is labeled as process 240 for distinction.

[0067] Process 240 includes step 241 of inputting the corrected medical image 135 into a machine learning method configured to generate a predicted noise image 245. The predicted noise image is an image containing the same number of pixels as the medical image and showing a measure of the estimated / predicted noise for that pixel (for each pixel in the medical image).

[0068] Next, process 240 performs step 242, which multiplies the predicted noise image 245 by a noise map. In other words, it renormalizes the estimated noise image 245 to generate a calibrated predicted noise image 247.

[0069] Naturally, if the medical image was not divided by the noise map in step 130, step 242 is modified to perform the reverse procedure performed in step 130, using the estimated noise image instead of the corrected noise image (i.e., step 242 performs the reverse procedure of step 130, using the estimated noise image 245 and noise map 107 as inputs). Thus, the estimated noise image can be calibrated by applying the opposite (or reverse) of the correction made to the medical image to generate the corrected medical image to the estimated noise image.

[0070] Next, in step 243, process 240 subtracts the calibrated predicted noise image from the medical image 105 to generate a denoised medical image 145.

[0071] In some implementations, calibrated predictive noise images are weighted (e.g., scaled down) before being subtracted from the medical images. This approach recognizes that some clinicians prefer to retain some (non-zero) levels of noise in the medical images to reduce the artificial appearance of denoised medical images (which can be distracting to clinicians if not addressed).

[0072] Therefore, step 243 includes subtracting a scaled version of the calibrated predictive noise image from the medical image. The scaled version can be calculated by multiplying the value of each pixel in the calibrated predictive noise image by a predetermined value. In this case, the predetermined value is between 0 and 1, for example, between 0.25 and 0.75.

[0073] In this way, process 240 is configured to receive a corrected medical image as input and provide a predicted noise image as output. The predicted noise image shows, for each pixel, the amount of noise predicted or estimated at that pixel. This may be on a pixel-by-pixel basis.

[0074] In the proposed approach, each machine learning method processes medical images that have already been normalized or homogenized (e.g., uncorrelated) based on statistical information about the noise. This means that the images provided as input to the machine learning method are already at a consistent and / or uncorrelated noise level. This allows the machine learning method to be trained using normalized medical images, reducing the risk of overfitting (e.g., to a specific reconstruction filter or noise level).

[0075] Returning to Figure 1, method 100 further includes step 150, which controls a user interface to provide a visual representation of the denoised medical image 145 output by process 140. The user interface is a display, such as a monitor.

[0076] In some embodiments, method 100 includes step 155 of saving the denoised medical image to, for example, memory or a storage unit. Step 155 may include saving the denoised medical image to the electronic medical record of the person for whom the denoised medical image was taken.

[0077] Figure 3 shows method 300 according to another embodiment.

[0078] Method 300 differs from Method 100 above in that step 110 for acquiring a medical image includes step 311 for acquiring a first medical image 305, step 312 for processing the first medical image using a frequency filter (filtering according to a predetermined frequency range) to acquire a first filtered medical image 315, and step 313 for setting the first filtered medical image as a medical image. The frequency filter is, for example, a high-pass filter or a band-pass filter.

[0079] In this way, the medical image that is denoised may be the frequency-filtered portion of the medical image.

[0080] The frequency-filtered portion is preferably the high-frequency portion of the medical image, that is, the part of the medical image that has frequencies higher than a predetermined frequency value. Here, it is recognized that noise in medical images is usually high-frequency, and that more efficient and improved noise reduction can be achieved by denoising only the high-frequency portion of the medical image.

[0081] In some embodiments, step 110 further includes step 314, which processes the first medical image using a different frequency filter to obtain a second filtered medical image 316. The second filtered medical image has a different frequency range than the first filtered medical image. In one example, the second filtered medical image is a portion of the (original) medical image that is not in a given frequency range.

[0082] For example, if the frequency filter is a high-pass filter, the second filtered medical image is the low-pass filtered portion of the medical image. In other words, the medical image is processed using a low-pass filter. Low-pass and high-pass filters can have the same cutoff frequency. Thus, the medical image is effectively divided (by steps 312 and 314) into a high-frequency medical image and a low-frequency medical image, forming the first filtered medical image and the second filtered medical image, respectively.

[0083] An alternative to the optional step 314 for generating a second filtered medical image 316 is to subtract the first filtered medical image 315 from the first medical image 305.

[0084] Similarly, if step 314 involves processing the first medical image using a filter such as a low-pass filter, step 312 may be modified to include subtracting the second filtered medical image 316 from the first medical image 305 to produce the first filtered medical image 315.

[0085] Method 300 may further include step 360 of combining the second filtered medical image 316 with the denoised medical image 145 output by process 140 in order to reshape the denoised version 349 of the first medical image. Step 360 may include simply summing the second filtered medical image with the denoised medical image output by process 140.

[0086] Similar to method 100, method 300 may include a step 150 of controlling a user interface to provide a visual representation of the denoised medical image 145 output by process 140, and / or a step 155 of saving the denoised medical image to, for example, memory or a storage unit. Method 300 is adapted accordingly.

[0087] The proposed method has been shown to be particularly effective when the machine learning method is a residual learning method. A residual learning method generates a predicted noise image containing the same number of pixels as the medical image as output, and provides a measurement of the predicted noise for that pixel (for each pixel).

[0088] Testing and analysis of the method proposed herein have shown that it sufficiently generalizes different reconstruction filters and also sufficiently generalizes higher noise levels than those observed during training of machine learning methods (e.g., those resulting from the use of low radiation levels). Therefore, the method proposed herein improves the robustness of practical machine learning.

[0089] It has been recognized that some methods for generating noise maps may result in calculated noise maps that do not accurately reflect the statistical variations in noise levels introduced by different reconstruction filters. Rather, they may reflect differences in noise levels due to other factors, such as widespread system errors or the use of low radiation doses.

[0090] In this case, in order to generalize the proposed method to different systems, the noise map obtained in step 120 of the method described above can be simply scaled by a global scaling factor before being used for processing medical images.

[0091] For example, in the field of CT imaging, it is well known that reconstruction filters typically consist of two parts or functions: a ramp function and an additional modulation transfer function (MTF). A more complete understanding is described in Thorsten M. Buzug (2008) "Computed Tomography" (Springer-Verlag, Berlin, Heidelberg). Generally, the various reconstruction filters used in CT imaging differ only in their MTF modulation portion.

[0092] This global scaling factor is the (Ramp*MTF) of the reconstruction filter used.2 It can be calculated from the area under the curve. MTF is the modulation transfer function. In fact, the noise dispersion is (Ramp * MTF) 2 It is thought to scale linearly with respect to the area below. The ratio (square root) of the area of ​​the reconstruction filter used to reconstruct the medical image during processing to the corresponding area of ​​the reconstruction filter used to train the machine learning method provides an appropriate scaling factor for the noise map.

[0093] For example, if noise map generation is image-based, or if reconstruction filters (used to generate medical images) are taken into consideration, scaling of the noise map is unnecessary.

[0094] In summary, this method allows us to use a pre-trained machine learning method for application to an unknown reconstruction filter by scaling the corresponding standard deviation noise map with the scaling coefficient described above. Since this coefficient is simply a function of the two filters, there is virtually no additional computational cost or overhead, and therefore no need to retrain the machine learning method.

[0095] Figure 4 illustrates the operation of the proposed concept for denoising medical images. Figure 4 provides two denoised CT head images, partially obscured (outside the area enclosed by the white circle in each image). Each CT image was generated by processing the same medical CT image acquired at a 25% dose level.

[0096] The first denoised CT head image 410 was generated using a conventional denoising machine learning method (particularly a convolutional neural network), and no normalization or modification of the medical image using a noise map, as proposed in this invention, was performed.

[0097] A second denoised CT head image 420 was generated using the proposed denoising method, specifically a denoising method in which the statistical parameters of the noise were normalized before processing using a machine learning method.

[0098] In both cases, the machine learning method was trained using CT images acquired at a 25% dose level and reconstructed using a first reconstruction filter. Medical CT images (which were later denoised to generate head image 420) were generated using a second, different reconstruction filter with less noise suppression. While different reconstruction filters have different noise characteristics, they also offer other advantages, such as providing sharper or smoother images and / or highlighting different anatomical features. Therefore, operators can choose noisier or less noisier reconstruction filters according to their clinical preferences.

[0099] The proposed method was found to generate denoised images with less noise, and this made the method more robust to reconstruction filters with higher noise levels than when the machine learning method used in the denoising process was being trained.

[0100] It should be noted that the noise map obtained during the generation of the second denoised CT head image was scaled by a factor of 1.4 according to the method described above, in order to more accurately reflect the higher noise levels in the image as a result of different reconstruction filters. The value of 1.4 was selected based on the mechanism for determining the scaling factor described above.

[0101] The proposed embodiment uses a machine learning method, which is an arbitrary self-training algorithm that processes input data to generate or predict output data. Here, the input data includes a modified medical image, and the output data includes either a denoised medical image or a predicted noisy image.

[0102] Suitable machine learning methods for using the present invention will be apparent to those skilled in the art. Examples of suitable machine learning methods include decision tree algorithms and artificial neural networks. Other machine learning methods such as logistic regression, support vector machines, or naive Bayes models are suitable alternatives.

[0103] The structure of an artificial neural network (or simply a neural network) is inspired by the human brain. A neural network consists of layers, each containing multiple neurons. Each neuron performs a mathematical operation. In particular, each neuron may contain different weightings of a single type of transformation (for example, the same type of transformation, such as a sigmoid, but with different weightings). In the input data processing process, the mathematical operations of each neuron are performed on the input data to produce a numerical output, and the output of each layer of the neural network is passed sequentially to the next layer, with the final layer providing the output.

[0104] The embodiments described herein are particularly advantageous when using residual learning machine learning methods such as residual neural networks. Unlike conventional neural networks, residual neural networks can use skip connections, and therefore, for example, the output of one layer can skip one or more layers (i.e., not all outputs of any given layer need to be provided consecutively as input to the next layer).

[0105] The training methods for machine learning methods are well known. Typically, such methods involve obtaining a training dataset containing training input data entries and corresponding training output data entries. The initialized machine learning method is applied to each input data entry to generate predicted output data entries. The machine learning method is then modified using the error between the predicted output data entries and the corresponding training output data entries. This process is repeated until the error converges, i.e., until the predicted output data entries are sufficiently similar to the training output data entries (e.g., ±1%). This is generally known as supervised learning.

[0106] For example, when a machine learning method is formed from a neural network, the mathematical operations (weighting) of each neuron are modified until the error converges. Known methods for modifying neural networks include gradient descent and backpropagation algorithms.

[0107] The training input data entries correspond to modified medical image examples. In particular, each training input data entry should include medical images that have been preprocessed or modified using a noise map (of the medical image). For example, it should include medical images that have gone through step 130, as described with reference to Method 100. This improves the reliability and accuracy when using the machine learning method in Method 100.

[0108] The training output data entries correspond to denoised medical images and / or noisy image examples. Information about the reconstruction filters used to generate the medical image examples for the training input data entries may be stored, for example, to facilitate the scaling of the noise map as described above.

[0109] For example, the training input data entries are medical images with added (corrected) artificial noise, while the training output data entries are medical images without added (corrected) artificial noise.

[0110] Any method proposed herein can be carried out by the imaging system itself (i.e., the system that generates the medical image, the processing system located on the same premises as the imaging system, a mobile device (such as a smartphone, tablet, or laptop), or by using a distributed processing system, i.e., the "cloud".

[0111] Those skilled in the art will be able to easily develop a processing system for carrying out the method described herein. Thus, each step in the flowchart represents a different action performed by the processing system, which can be performed by the corresponding module of the processing system.

[0112] Therefore, embodiments utilize a processing system. A processing system can be implemented in various ways using software and hardware to perform a variety of required functions. A processor is an example of a processing system employing one or more microprocessors programmed using software (e.g., microcode) to perform the required functions. However, a processing system can be implemented regardless of the processor used, and can also be implemented as a combination of dedicated hardware for performing some functions and processors (e.g., one or more programmed microprocessors and associated circuits) for performing other functions.

[0113] Examples of processing system components that may be used in various embodiments of this disclosure include, but are not limited to, conventional microprocessors, application-specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs).

[0114] In various implementations, a processor or processing system may be associated with one or more storage media, such as volatile and non-volatile computer memory, including RAM, PROM, EPROM, and EEPROM. The storage media may be encoded with one or more programs that perform the necessary functions when executed on one or more processors and / or processing systems. The various storage media may be fixed within the processor or processing system, or they may be transportable so that the one or more programs stored therein can be loaded into the processor.

[0115] As a further embodiment, Figure 5 shows an example of a processing system 500 in which one or more parts of the embodiment can be used. The functions of the processing system 500 are utilized for the various operations described above. For example, one or more parts of a system for denoising medical images can be incorporated into any element, module, application, and / or component described herein. In this regard, it should be understood that the functional blocks of the system can run on a single computer or be distributed across several computers and locations (e.g., connected via the Internet).

[0116] The processing system 500 includes, but is not limited to, PCs, workstations, laptops, PDAs, palm devices, servers, and storage devices. Generally, with respect to the hardware architecture, the processing system 500 includes one or more processors 501, memory 502, and one or more I / O devices 507 that are communicatively coupled via a local interface (not shown). The local interface is, for example, one or more buses, or other wired or wireless connections as known in the art, but is not limited to these. The local interface may have controllers, buffers (caches), drivers, repeaters, and receivers to enable communication. Furthermore, the local interface may include addresses, control units, and / or data connections to enable proper communication between the aforementioned components.

[0117] The processor 501 is a hardware device that executes software which may be stored in memory 502. The processor 501 may be substantially any custom-made or commercially available processor, a central processing unit (CPU), a digital signal processor (DSP), or an auxiliary processor from among multiple processors associated with the processing system 500, and the processor 501 may also be a semiconductor-based microprocessor (in the form of a microchip) or a microprocessor.

[0118] Memory 502 may include one or a combination of volatile memory elements (random access memory (RAM) such as dynamic random access memory (DRAM) and static random access memory (SRAM)) and non-volatile memory elements (such as ROM, erasable programmable read-only memory (EPROM), electronically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), tape, compact disc read-only memory (CD-ROM), disk, floppy disk, cartridge, cassette, etc.). Furthermore, memory 502 may incorporate electronic media, magnetic media, optical media, and / or other types of storage media. Memory 502 may have a distributed architecture in which various components are located in separate locations but are accessible by the processor 501.

[0119] The software in memory 502 may include one or more separate programs, each containing an ordered list of executable instructions for performing a logical function. The software in memory 502 may also include, according to an exemplary embodiment, a suitable operating system (O / S) 505, a compiler 504, source code 503, and one or more applications 506. As shown in the figure, application 506 includes a number of functional components for performing the features and operations of the exemplary embodiment. Application 506 of the processing system 500 may represent various applications, computing units, logic, functional units, processes, operations, virtual entities, and / or modules according to an exemplary embodiment, but application 506 is not intended to be limiting.

[0120] The operating system 505 controls the execution of other computer programs and provides scheduling, input / output control, file and data management, memory management, and communication control, as well as related services. The inventor intends that application 506 for carrying out exemplary embodiments is applicable to any commercially available operating system.

[0121] Application 506 can be a source program, an executable program (object code), a script, or any other entity including a set of instructions to be executed. If it is a source program, the program is usually translated through a compiler (such as Compiler 504), assembler, interpreter, etc., which may or may not be included in memory 502, so that it can function properly in relation to O / S 505. Furthermore, Application 506 can be written as an object-oriented programming language with classes of data and methods, or as a procedural programming language with routines, subroutines, and / or functions (e.g., C, C++, C#, Pascal, BASIC, API calls, HTML, XHTML, XML, ASP scripts, JavaScript, FORTRAN, COBOL, Perl, Java, ADA, .NET, etc., but not limited to these).

[0122] I / O devices 507 include input devices such as, but are not limited to, a mouse, keyboard, scanner, microphone, and camera. Furthermore, I / O devices 507 include output devices such as, but are not limited to, a printer and display. Finally, I / O devices 507 further include devices that communicate both input and output, such as, but are not limited to, a NIC or modulator / demodulator (for accessing remote devices, other files, devices, systems, or networks), radio frequency (RF) or other transceivers, telephone interfaces, bridges, and routers. I / O devices 507 also include components for communication over various networks, such as the Internet and intranets.

[0123] If the processing system 500 is a PC, workstation, intelligent device, etc., the software in memory 502 may also include a Basic Input / Output System (BIOS) (omitted for simplicity). The BIOS is a set of essential software routines that initialize and test the hardware at startup, boot the OS 505, and support data transfer between hardware devices. The BIOS is stored in some type of read-only memory, such as ROM, PROM, EPROM, or EEPROM, so that it is executed when the processing system 500 is started.

[0124] During the operation of the processing system 500, the processor 501 is configured to execute software stored in memory 502, exchange data with memory 502, and generally control the operation of the processing system 500 according to the software. The application 506 and the OS 505, in whole or in part, are read by the processor 501, and in some cases, buffered within the processor 501 before being executed.

[0125] If Application 506 is implemented in software, it should be noted that Application 506 may be stored in substantially any computer-readable medium for use in or in connection with any computer-related system or method. In the context of this document, computer-readable medium may be an electronic, magnetic, optical, or other physical device or means capable of storing or preserving a computer program for use in or in connection with a computer-related system or method.

[0126] Application 506 can be embodied in any computer-readable medium for use in or in connection with an instruction execution system, apparatus, or device (such as a computer-based system, a processor-based system, or any other system capable of fetching instructions from an instruction execution system, apparatus, or device and executing those instructions). In the context of this document, “computer-readable medium” can be any means by which a program can be stored, communicated, propagated, or transferred for use in or in connection with an instruction execution system, apparatus, or device. Computer-readable mediums include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, devices, or propagation media.

[0127] Figure 6 schematically shows system 600, which includes an imaging system 602 and a processing system 612. Here, the imaging system is a CT scanner configured for spectral (multi-energy) imaging. However, other forms of imaging systems can be used. The processing system 612 may be embodied as the processing system 500 described above.

[0128] The illustrated imaging system 602 typically includes a fixed gantry 604 and a rotating gantry 606. The rotating gantry is rotatably supported by the fixed gantry 604 and rotates around the examination area 608 about the z-axis. A subject support 610, such as a couch, supports an object or subject within the examination area 608.

[0129] A radiation source 612, such as an X-ray tube, is rotatably supported by a rotating gantry 606 and rotates with the rotating gantry 606, emitting radiation across the inspection area 608. In one example, the radiation source 612 includes a single broad-spectrum X-ray tube. In another example, the radiation source 612 includes a single X-ray tube configured to switch between at least two different emission voltages (e.g., 80 kVp and 640 kVp) during a scan. In yet another example, the radiation source 612 includes two or more X-ray tubes configured to emit radiation with different average spectra. In yet another example, the radiation source 612 includes a combination of these.

[0130] The radiation-sensitive detector array 614 defines the arc range for angles opposite the radiation source 612 across the inspection area 608. The radiation-sensitive detector array 614 detects radiation crossing the inspection area 608 and generates an electrical signal (projection data) indicating it. While the radiation source 612 includes a single broad-spectrum X-ray tube, the radiation-sensitive detector array 612 includes energy-resolved detectors (e.g., direct conversion photon count detectors, at least two sets of scintillators (multilayer) with different spectral sensitivities). In kVp switching and multi-tube configurations, the detector array 614 may include single-layer detectors, direct conversion photon count detectors, and / or multilayer detectors. Direct conversion photon count detectors include conversion materials such as CdTe, CdZnTe, Si, Ge, GaAs, or other direct conversion materials. Examples of multilayer detectors include double-decker detectors such as the one described in U.S. Patent No. 7,968,853(B2), filed April 60, 2006, titled "Double Decker Detector for Spectral CT." This patent is incorporated herein by reference in its entirety.

[0131] The reconstructor 616 receives spectral projection data from the detector array 614 and reconstructs spectral volumetric image data, such as sCCTA image data, high-energy images, low-energy images, photoelectric images, Compton scattering images, iodine images, calcium images, virtual non-contrast images, bone images, soft tissue images, and / or other base material images. The reconstructor 616 can also reconstruct non-spectral volumetric image data, such as by combining spectral projection data and / or spectral volumetric image data. Generally, spectral projection data and / or spectral volumetric image data include data for at least two different energies and / or energy ranges.

[0132] In this way, the reconstructor 616 generates or reconstructs a medical image.

[0133] Here, the processing system 618 functions as an operator console. The console 618 includes human-readable output devices such as a monitor and input devices such as a keyboard and mouse. Software residing in the console 618 allows the operator to interact with the scanner 602 and operate the scanner 102 via a graphical user interface (GUI) or the like. The console 618 further includes a processor 620 (e.g., a microprocessor, controller, central processing unit) and a computer-readable storage medium 622 (excluding non-temporary media), as well as temporary media such as a physical memory device. The computer-readable storage medium 622 contains instructions 624 for denoising the generated medical images; in other words, it contains a medical image denoising unit 625. The processor 620 is configured to execute instructions 624. The processor 620 may be additionally configured to execute one or more computer-readable instructions carried by carrier waves, signals, and / or other temporary media. In a modified example, the processor 620 and the computer-readable memory medium 622 are part of a separate processing system distinct from the processing system 618.

[0134] It will be understood that the disclosed methods are preferably computer implementations. Therefore, the concept of a computer program is also proposed, which includes code means for implementing any described method when executed on a processing system such as a computer. Thus, different parts, lines, or blocks of code in a computer program according to one embodiment may be executed by a processing system or computer to perform any method described herein. In some alternative implementations, the functions shown in a block diagram or flowchart may occur in a different order than shown in the diagram. For example, two blocks shown consecutively may actually be executed substantially simultaneously, or these blocks may be executed in reverse order depending on the functions involved.

[0135] Modifications of the disclosed embodiments can be understood and implemented by those skilled in the art in carrying out the claimed invention, based on a review of the drawings, disclosures, and appended claims. In the claims, the term “equipped with” is not exclusive to other elements or steps, and singular elements are not exclusive to plural elements. A single processor or other unit may perform the functions of several items described in the claims. The mere fact that certain means are described in different dependent claims does not mean that combinations of these means cannot be used advantageously. Where a computer program is described above, it may be stored / distributed on any suitable medium, such as optical storage media or solid-state media, supplied together with or as part of other hardware, but it may also be distributed in other forms, such as via the Internet or other wired or wireless communication systems. Note that where the term “adapted to be” is used in the claims or description, the term “adapted to be” is intended to be equivalent to the term “configured to be.” Any reference numerals in the claims should not be construed as limiting the scope.

Claims

1. A computer method for denoising medical images and generating denoised medical images, The steps include acquiring the medical image formed from multiple pixels, The steps include obtaining a noise map that includes estimated measurements of the statistical parameters of the noise of each pixel in the medical image, The steps include: modifying the medical image using the noise map in order to generate a corrected medical image; The steps include: processing the modified medical image using a machine learning method to generate the denoised medical image; Includes, The step of correcting the medical image includes the step of dividing the medical image by the noise map. Computerized implementation method.

2. The step of processing the corrected medical image is: A step of processing the corrected medical image using the machine learning method in order to generate a predictive noise image, wherein the predictive noise image represents the amount of noise predicted at each pixel of the corrected medical image. To generate a calibrated predictive noise image, the steps include multiplying the corrected medical image by the noise map, To generate the denoised medical image, the steps include subtracting the calibrated predictive noise image or a scaled version of the calibrated predictive noise image from the medical image, The computer implementation method according to claim 1, including the method described in claim 1.

3. The computer implementation method according to claim 1, wherein the step of processing the modified medical image includes inputting the modified medical image into the machine learning method and receiving the denoised medical image as output from the machine learning method.

4. The computer implementation method according to claim 1, wherein the machine learning method includes a neural network.

5. The computer implementation method according to claim 1, wherein the noise map provides an estimate of the standard deviation or variance of the noise of each pixel of the medical image.

6. The computer implementation method according to claim 1, wherein the noise map provides an estimated correlation between the noise of each pixel in the medical image and the noise of one or more adjacent pixels.

7. The aforementioned medical image is one of several medical images that represent the same scene and are generated by a multi-channel imaging process. The computer implementation method according to claim 1, wherein the noise map provides an estimated value of the covariance or correlation between the noise of each pixel in the medical image and the noise of a corresponding pixel in another medical image among the plurality of medical images.

8. The step of acquiring the aforementioned medical image is: The first step is to acquire a medical image, The steps include processing a first filtered medical image using a frequency filter to obtain a first filtered medical image having values ​​within a predetermined frequency range, The steps include setting the first filtered medical image as the medical image and The computer implementation method according to claim 1, including the method described in claim 1.

9. The steps include processing the first medical image to obtain a second filtered medical image having a second value within a different predetermined frequency range, The steps include combining the second filtered medical image and the denoised medical image to generate a denoised first medical image, and The computer implementation method according to claim 8, further comprising:

10. The aforementioned medical image is a medical image reconstructed from raw data using a first reconstruction algorithm. The computer implementation method according to claim 1, wherein the machine learning method is a machine learning method trained using a training dataset which includes one or more training images reconstructed from raw data using a second different reconstruction algorithm.

11. The computer implementation method according to claim 1, wherein the medical image is a computed tomography medical image.

12. A computer program comprising computer program code means, wherein the computer program code means, when executed on a computing device having a processing system, causes the processing system to perform all the steps of the computer implementation method described in claim 1.

13. A processing system for denoising medical images and generating denoised medical images, wherein the processing system is: The medical image formed from multiple pixels is acquired, A noise map is obtained that includes estimated measurements of the statistical parameters of the noise of each pixel in the aforementioned medical image. To generate a corrected medical image, the medical image is modified using the noise map, To generate the denoised medical image, the modified medical image is processed using a machine learning method. It is configured in such a way, The step of correcting the medical image includes the step of dividing the medical image by the noise map. Processing system.

14. The processing system according to claim 13, A medical imaging system that generates the medical image and provides the generated medical image to the processing system. A system equipped with these features.

Images