Evaluating medical image quality

Abstract

Embodiments of the present invention aim to provide a method for evaluating the quality of medical images. This is achieved by using a machine learning model to determine absolute image quality values ​​and comparative image quality values ​​of the medical image. The comparative image quality values ​​describe the relative quality of the image relative to an appropriately selected reference image. Then, an overall image quality value is determined based on the absolute image quality values ​​and the comparative image quality values.

Description

Technical Field

[0001] This invention relates to the field of imaging, and more specifically to the field of evaluating the quality of medical images. Background Technology

[0002] In medical imaging examinations, accurately assessing image quality and applicability is crucial. If the image quality obtained during a medical examination is insufficient, key measurements and diagnoses may be impossible, and the medical examination may be considered a failure.

[0003] However, image quality is often difficult to assess and is highly subjective. For patients with technical difficulties, it is often challenging to determine whether the image quality acquired during a medical examination is sufficient for diagnostic and examination procedures, or whether the medical professional performing the patient's examination should attempt to acquire more images in hopes of obtaining better quality images.

[0004] Machine learning models can be used to automate image quality scoring, but image quality and suitability scores are often subjective, making it extremely challenging to label appropriate data for training such models.

[0005] In their paper "Generalization evaluation of numerical observers for image quality assessment," published at the 2006 IEEE NucleScience Symposium Conference Record (October 29, 2006, pp. 1696-1698, XP093174795), Jovan G. Brankov et al. presented a numerical observer (NO) using channelized support vector machines (CSVM) for a lesion detection task, comparing its performance with the widely used channelized Hotelling observer (CHO). Cross-validation was used to quantify the generalization ability of NO, as this allows us to evaluate how NO performs when trained on one set of images but tested on another.

[0006] In their paper "Artifact- and content-specific quality assessment for MRI with image rulers by K," published in *Medical Image Analysis* (Oxford, Oxford University Press, Vol. 77, January 20, 2022, XP086982022), Lei et al. proposed a framework with a multi-task convolutional neural network (CNN) model. This model is trained with calibrated labels and inferred using image rulers, where the manually calibrated labels follow an explicit and effective annotation task. The image rulers address different quality standards and provide a specific way to interpret the raw scores of the CNN output. In particular, the model disclosed in this paper supports the evaluation of noise and motion artifacts in MRI images. Summary of the Invention

[0007] This invention is defined by the claims.

[0008] According to an example of one aspect of the present invention, a computer-implemented method for evaluating the quality of a medical image is provided. The method includes: processing the medical image using a first machine learning model to determine an absolute image quality value of the medical image; analyzing the medical image to select a reference image from a plurality of different possible reference images; processing the medical image and the selected reference image using a second machine learning model to determine a comparative image quality value describing the quality of the medical image relative to the selected reference image; and determining an overall image quality value of the medical image based on the absolute image quality value and the comparative image quality value.

[0009] Therefore, the proposed concepts aim to provide schemes, solutions, ideas, designs, methods, and systems related to evaluating the quality of medical images. Specifically, the embodiments aim to provide a computer-implemented method for automatically evaluating the quality of medical images by calculating absolute image quality values ​​and comparing image quality values, wherein the compared image quality describes the quality of the image relative to an appropriately selected reference image.

[0010] Specifically, it is proposed that both absolute image quality values ​​and comparative image quality values ​​can be used to evaluate the overall image quality of medical images. The second value is obtained by feeding the medical image to be evaluated and a pre-selected reference image as input to a machine learning model, which then outputs the relative quality of the medical image with respect to the reference image. The reference image is selected from multiple different possible reference images based on the analysis of the medical image to be evaluated.

[0011] Automated image quality assessment is of great importance in the field of medical imaging. During medical imaging examinations, ensuring a sufficient number of high-quality images are acquired for measurement and diagnosis is crucial. Furthermore, in the diagnostic phase, automated medical image quality assessment is valuable for comparing the applicability of different images of the same patient within the diagnostic process.

[0012] It has been recognized that comparative image quality values ​​obtained through pairwise comparisons between images are generally more consistent and reliable than absolute image quality values ​​obtained by machine learning models trained on image datasets based on subjectively labeled absolute image quality values. Furthermore, in the context of medical imaging, comparative image quality values ​​are particularly helpful in indicating whether a given medical image has sufficient quality for a reliable image diagnosis, as the image quality can be compared to images from which previous diagnoses have been successfully performed. It is also recognized that image quality assessment can depend on several characteristics of the image being evaluated; therefore, appropriate reference images should be selected based on the analysis of the medical image being evaluated to ensure the acquisition of the most suitable comparative image quality value.

[0013] Therefore, the proposed invention utilizes both absolute image quality values ​​and comparative image quality values ​​to automatically provide an overall image quality value describing the image quality of medical images. Compared to previously known manual or automated image quality assessments based on subjective annotation of absolute image quality scores, the resulting image quality assessment is likely to be more reliable and accurate.

[0014] In some embodiments, analyzing the medical image to select a reference image may further include: processing the medical image using an image recognition model to determine at least one image parameter of the medical image; and selecting a reference image from a plurality of different possible reference images based on the determined at least one image parameter. In this way, a suitable reference image can be selected based on a variety of parameters that may affect the image quality of the medical image to be evaluated.

[0015] In some embodiments, the at least one image parameter may include at least one of the following: view position; anatomical features; orientation of anatomical features; patient identification value; examination identification value; date; and image sensor identification value. Thus, for example, a reference image may be selected to show the same anatomical features as the medical image or anatomical features specific to the same patient. Image quality can vary depending on various parameters of the medical image, therefore selecting a suitable reference image based on at least one of these parameters may be important to ensure the accuracy of the determined comparative image quality values.

[0016] In some embodiments, the first machine learning model may be trained using a deep learning algorithm configured to receive an array of training inputs and corresponding known outputs. A training input to the second machine learning model may be a first training image in a pair of training images, and the corresponding known output may be an absolute image quality value, wherein the absolute image quality value is calculated based on comparative training image quality values ​​associated with the pair of training images. The comparative training image quality values ​​describe the quality of the first training image relative to the second training image in the pair of training images based on annotations of trained individuals. Calculating the absolute quality value based on the comparative quality value of the training dataset in this way can be more accurate than directly obtaining the absolute quality value (which is typically subjective).

[0017] In some embodiments, calculating the absolute quality value of the first training image based on the comparative training image quality values ​​associated with the pair of training images includes using an Elo-based transformation system. This provides an accurate, robust, and efficient method for obtaining absolute image quality values ​​from pairwise image quality comparison values.

[0018] In some embodiments, the absolute image quality value may include multiple values ​​describing the corresponding image quality of several different anatomical features present in the medical image. Therefore, the proposed method can provide detailed feedback on the quality of the various features depicted in the image, which can provide useful information for medical professionals assessing the image's diagnostic usefulness, as a diagnosis may only require high-quality imaging of specific anatomical features.

[0019] In some embodiments, the selected reference image may have a substantially identical view orientation to the medical image to be evaluated. This is a useful feature because image quality can vary considerably with view orientation, and therefore, by ensuring that the selected reference image has the same view orientation as the medical image to be evaluated, the image quality between the two images can be compared more fairly.

[0020] In some embodiments, the selected reference image depicts the same anatomical features as the medical image to be evaluated. This allows for a direct comparison of the image quality of the depicted anatomical features between the medical image and the reference image, thereby ensuring more accurate and detailed comparison quality values.

[0021] In some embodiments, the comparative image quality values ​​may include multiple values ​​describing the relative image quality of each anatomical feature present in the medical image to be evaluated compared to the image quality of the same anatomical feature depicted in the reference image. In this way, the proposed method can provide detailed information on the comparative image quality of each anatomical feature in a medical image. This can be useful in medical imaging, as only a portion of the entire system being imaged may be important for diagnosis and evaluation; therefore, a detailed breakdown of the image quality of each feature in the image enables medical professionals to determine whether the medical image is suitable for its intended use.

[0022] In some embodiments, the method may further include determining a suitability value for the medical image based on comparing a determined overall quality value with a first threshold. In this way, the overall quality value can be used to automatically determine whether the medical image is suitable for use in a diagnostic process or for further assessment of the patient's condition.

[0023] In some embodiments, the overall quality value may include at least one of the following: the absolute image quality value; and the comparative image quality value. In this way, the overall image quality value may simply include one or both of the determined image quality values. For example, the method may further include determining which image quality value might be more appropriate and / or more accurate, and the overall image quality value may be selected from the absolute image quality value and the comparative image quality value based on this determination.

[0024] According to another aspect of the present invention, a computer program including code modules is provided, wherein when the program is run on a processing system, the code modules are used to implement the methods of any of the proposed embodiments.

[0025] According to another aspect of the present invention, a processing apparatus is provided, configured to: process a medical image using a first machine learning model to determine an absolute image quality value of the medical image; analyze the medical image to select a reference image from a plurality of different possible reference images; process the medical image and the selected reference image using a second machine learning model to determine a comparative image quality value describing the quality of the medical image relative to the selected reference image; and determine an overall image quality value of the medical image based on the absolute image quality value and the comparative image quality value.

[0026] According to another aspect of the present invention, a medical imaging system suitable for evaluating the quality of medical images is provided. The system includes a processing device configured to: process the medical image using a first machine learning model to determine an absolute image quality value of the medical image; analyze the medical image to select a reference image from a plurality of different possible reference images; process the medical image and the selected reference image using a second machine learning model to determine a comparative image quality value describing the quality of the medical image relative to the selected reference image; and determine an overall image quality value of the medical image based on the absolute image quality value and the comparative image quality value.

[0027] In an embodiment, the medical imaging system may further include at least one of the following: an image sensor; and an output interface configured to output an indication of the overall image quality of the determined medical image.

[0028] The embodiments can be used in conjunction with conventional / existing medical imaging methods and systems. In this way, the embodiments can be integrated into conventional systems to improve and / or expand their functionality and capabilities. Therefore, the proposed embodiments can provide an improved medical imaging system capable of automatically assessing image quality.

[0029] Therefore, it is possible to propose a concept for evaluating the quality of medical images by simultaneously using absolute image quality values ​​and comparative image quality values. These and other aspects of the invention will become apparent and will be illustrated with reference to one or more embodiments described below. Attached Figure Description

[0030] To better understand the invention and to more clearly illustrate how the invention can be implemented, reference will now be made to the accompanying drawings by way of example only, in which: Figure 1 This is a simplified flowchart of a method for evaluating the quality of medical images according to the proposed embodiments; Figure 2 This is a simplified diagram of the display interface used during the pairwise labeling process of the training dataset; Figure 3 It is a simplified diagram of the labeled network generated by pairwise annotations of the training dataset; Figure 4 This is a simplified block diagram of a system for evaluating the quality of medical images according to the proposed embodiments; Figure 5 An example output interface of a system for evaluating the quality of medical images according to a proposed embodiment is described; and Figure 6 An example of a computer that can employ one or more parts of an embodiment is shown. Detailed Implementation

[0031] The present invention will now be described with reference to the accompanying drawings.

[0032] It should be understood that while the detailed description and specific examples indicate exemplary embodiments of the apparatus, system, and method, they are for illustrative purposes only and are 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 become better understood from the following description, claims, and drawings. It should be understood that the drawings are schematic only and are not drawn to scale. It should also be understood that the same reference numerals are used in all drawings to indicate the same or similar parts.

[0033] Embodiments of this disclosure relate to various techniques, methods, schemes, and / or solutions related to evaluating the image quality of medical images. Based on the proposed concepts, multiple possible solutions can be implemented individually or in combination. That is, although these possible solutions may be described separately below, two or more of these possible solutions may be implemented in one or another combination.

[0034] Embodiments of the present invention aim to provide a computer-implemented method for evaluating the quality of medical images. This is achieved by using a machine learning model to determine absolute image quality values ​​and comparative image quality values ​​of the medical image. The comparative image quality values ​​describe the relative quality of the image relative to an appropriately selected reference image. Then, an overall image quality value is determined based on the absolute image quality values ​​and the comparative image quality values.

[0035] Therefore, the proposed concept aims to provide schemes, solutions, concepts, designs, methods, and systems related to the automatic assessment of image quality using both absolute and relative quality scores to describe image quality and its potential uses in diagnosis.

[0036] Automated image quality scoring in this manner during medical examinations helps overcome examination failures caused by insufficient image quality. By applying the method described herein, healthcare professionals can be provided with real-time automated quality assessment of images acquired during examinations. Previously, healthcare professionals might acquire more images than necessary to ensure that at least some of the acquired images are of sufficient diagnostic quality. The proposed method enhances healthcare professionals' confidence in the quality of the acquired images. This allows for the acquisition of fewer images during the examination, thus shortening the examination time. Furthermore, examination failures can be prevented if the performing healthcare professional is notified early on when a sufficient number of high-quality images are acquired.

[0037] Accurate image quality assessment can be extremely useful when using medical images for measurement or diagnosis. If an image has low quality, it may be suitable for some purposes but not others. Therefore, automated image quality assessment at this stage allows for the selection of the optimal image for certain measurement and / or diagnostic procedures. For example, some images may have very high image quality for one feature of the imaged system, but lower quality for other features. This might limit the image to diagnosing only conditions related to those features that affect the high image quality.

[0038] In summary, the following steps outline the process based on the proposed concept(s): Step 1: Determine the absolute image quality value of the medical image to be evaluated. This is achieved by processing the medical image using a first machine learning model.

[0039] Step 2: Based on the analysis of the medical image to be evaluated, select a suitable reference image. In some embodiments, a reference image that has the same view position as the medical image or depicts the same anatomical features can be selected.

[0040] Step 3: Using a second machine learning model, determine a comparative image quality value that describes the quality of the medical image to be evaluated relative to the quality of a selected reference image.

[0041] Step 4: Determine the overall image quality value based on the determined absolute image quality value and the determined comparative image quality value.

[0042] Image quality metrics relate to the fidelity of an image to the anatomical features it depicts. Therefore, image quality values ​​can include metrics related to resolution, signal-to-noise ratio, dynamic range, contrast, aberrations, and the presence of artifacts. Other metrics related to image accuracy and sharpness can also be included in the image quality value. Importantly for this invention, image quality values ​​can include higher-level image quality values ​​that are particularly relevant to the visibility of the anatomical features depicted in the image. For example, for images of the cardiovascular system, image quality values ​​could describe the visibility of vessel wall segments or valves. The specific information included in the image quality value can depend on the type of imaging technique used to acquire the image. Image quality values ​​can be qualitative or quantitative. For example, qualitative image quality values ​​can be high, medium, or low, while quantitative image quality values ​​can be quality level values, for example, between 1 and 100.

[0043] The medical images to be evaluated in this invention can be any medical image requiring image quality assessment. For example, the medical image can be an image of the anatomical features of a subject (human or animal) acquired using any available medical imaging technique. This can include X-ray images, ultrasound images, CT (X-ray computed tomography) images, PET (positron emission tomography) scanner images, MRI (magnetic resonance imaging) images, or ultrasound images.

[0044] The method of this invention relies on the use of machine learning models. A first machine learning model and a second machine learning model employ machine learning algorithms to determine absolute image quality values ​​and comparative image quality values, respectively. A machine learning algorithm is any self-trained algorithm that processes input data to produce or predict output data. Here, for the first machine learning model, the input data includes the medical image to be evaluated; for the second machine learning model, the input data includes the medical image and a selected reference image. For the first machine learning model, the output data includes the absolute image quality value of the medical image; for the second machine learning model, the output data includes comparative image quality values ​​describing the quality of the medical image relative to the selected reference image.

[0045] The machine learning algorithms applicable to this invention will be readily apparent to those skilled in the art. Examples of suitable machine learning algorithms include decision tree algorithms and artificial neural networks. Other machine learning algorithms such as logistic regression, support vector machines, or Naive Bayes models are also suitable alternatives. By way of example only, one or more machine learning models may employ convolutional neural networks (CNNs), as CNNs have proven particularly successful in analyzing images and are able to identify objects within images with a much lower error rate than other types of neural networks.

[0046] Artificial neural networks (or simply neural networks) are inspired by the human brain. A neural network consists of multiple layers, each containing multiple neurons. Each neuron performs a mathematical operation. Specifically, each neuron can contain different weighted combinations of a single type of transformation (e.g., transformations of the same type, such as sigmoid, but with different weights). During the processing of input data, the mathematical operation of each neuron is performed on the input data to produce a numerical output, and the output of each layer in the neural network is sequentially fed to the next layer. The final layer provides the output.

[0047] There are various types of neural networks, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs). Some embodiments of the present invention may employ CNN-based learning algorithms because CNNs have proven particularly successful in analyzing images and are able to identify objects within images with a much lower error rate than other types of neural networks.

[0048] CNNs typically consist of multiple layers, including convolutional layers, pooling layers, and fully connected layers. Convolutional layers consist of a set of learnable filters that extract features from the input. Pooling layers are a form of non-linear downsampling that reduces the amount of data by combining the outputs of multiple neurons in one layer into a single neuron in the next layer. Fully connected layers connect each neuron in one layer to all neurons in the next layer.

[0049] Alternatively, embodiments may employ a Visual Transformer (ViT). A Visual Transformer is a deep learning algorithm that relies on analyzing pixel patches in an image. Visual Transformers have proven as successful as CNNs on image classification tasks while significantly reducing the computational resources required for training.

[0050] Methods for training machine learning algorithms are well-known. Typically, such methods involve obtaining a training dataset that includes training input data entries and corresponding training output data entries. An initialized machine learning algorithm is applied to each input data entry to generate a predicted output data entry. The error between the predicted output data entry and the corresponding training output data entry is used to modify the machine learning algorithm. This process can be repeated until the error converges and the predicted output data entry is sufficiently similar to the training output data entry (e.g., within ±1%). This is often referred to as a supervised learning technique.

[0051] For example, the weights of the mathematical operations for each neuron can be modified until the error converges. Known methods for modifying neural networks include gradient descent, backpropagation, and other algorithms.

[0052] In other words, the first and second machine learning models can be trained using a deep learning algorithm configured to receive an array of training inputs and corresponding known outputs. The training input for the first machine learning model is medical images, and the corresponding known output is the absolute image quality value for each medical image. The training input for the second machine learning model is a pair of medical images comprising the first and second training images, and the corresponding known output is a comparative training image quality value describing the quality of the first training image relative to the second training image. In this way, the first machine learning model can be trained to output an absolute image quality value describing the image quality of the medical image to be evaluated, while the second machine learning model can be trained to output a comparative image quality value describing the quality of the medical image relative to a reference image.

[0053] Now for reference Figure 1 A simplified flowchart of a method 100 for evaluating the quality of medical images according to the proposed embodiments is shown.

[0054] The method begins with step 110, which determines an absolute image quality value for a medical image. This step is achieved by processing the medical image using a first machine learning model. In this exemplary embodiment, the determined absolute image quality includes multiple values ​​describing the corresponding image quality of various anatomical features present in the medical image. This can be advantageous because image quality can vary significantly across the field of view and for features with different orientations relative to the image's viewing direction. For diagnostic and measurement purposes, only certain features depicted in the image may require high image quality, so this information can be useful in determining whether the medical image is suitable for use. Furthermore, when the invention is used during an examination, this detailed information can help guide the medical professional performing the examination on what adjustments need to be made to the imaging process to ensure an image of sufficient quality. In this exemplary embodiment, the absolute image quality value associated with each feature in the image is determined to be high, medium, or low.

[0055] The method further includes step 120. In this step, the medical image is analyzed to select a reference image from a plurality of different possible reference images. In this exemplary embodiment, step 120 further includes sub-steps 122 and 124. Step 122 includes processing the medical image using an image recognition model to determine at least one image parameter of the medical image. In this exemplary embodiment, the determined at least one image parameter describes anatomical features present in the medical image, the view position associated with the medical image, and the imaging technique used to obtain the image. For example, the image recognition model may determine that the medical image is a posteroanterior view ultrasound image. Step 124 includes selecting a reference image from a plurality of possible reference images based on the determined at least one image parameter. For example, for the above-described medical image, the reference image may be selected as a posteroanterior view ultrasound image also imaged in a posteroanterior view position.

[0056] Once a reference image is selected, the method proceeds to step 130, which involves processing the medical image and the reference image using a second machine learning model to determine comparative image quality values ​​describing the quality of the medical image relative to the quality of the selected reference image. In this exemplary embodiment, the comparative image quality values ​​include multiple values ​​describing the relative image quality of each anatomical feature present in the medical image compared to the image quality of the same anatomical feature depicted in the reference image. In this exemplary embodiment, this includes making a series of determinations, relative to the reference image, that the image quality in the medical image is higher, lower, or equal for each anatomical feature in the medical image. In this way, information is provided for a direct comparison of the image quality of features coexisting in both the reference image and the medical image. This can be useful when the method is applied to a medical examination procedure, as the medical professional performing the examination can obtain information describing whether a particular feature was imaged at a higher quality.

[0057] Once the comparative image quality and the absolute image quality are determined, the method proceeds to step 140, which includes determining an overall image quality value based on the absolute image quality value and the comparative image quality value. In this exemplary embodiment, the overall image quality includes at least one of the following: the absolute image quality value and the comparative image quality value.

[0058] The method then proceeds to step 150, which involves determining a suitability value for the medical image based on the overall image quality value determined in step 140. This value, based on the image quality of the medical image, describes how suitable the medical image is for diagnostic and further evaluation processes. For example, if the image quality is considered low, the medical image may be considered unsuitable for diagnosis and measurement. The suitability value can be a binary judgment of "suitable" or "not suitable," or it can provide a more detailed and nuanced judgment about how useful the image might be.

[0059] Although in this exemplary embodiment, the absolute image quality value and the comparative image quality value include multiple values ​​related to the image quality of various anatomical features depicted in the medical image, this is not necessarily the case in other embodiments. For example, in an alternative embodiment, the absolute image quality value and the comparative image quality value may simply be a single value describing the image quality of the entire image. Similarly, although in the above example, the absolute image quality value and the comparative image quality value are qualitative values, they may also be quantitative in other examples. For example, the absolute image quality value may be an image quality level value between 1 and 10, while the comparative image quality value may be a difference in the quality of the medical image relative to the quality of a reference image on the same level scale.

[0060] Similarly, other aspects of the above method can be implemented differently in alternative embodiments. For example, although in the above method, at least one image parameter determined by the image recognition model relates to anatomical features present in the medical image, the view orientation associated with the medical image, and the imaging technique used to acquire the medical image, different image parameters can be determined in other embodiments. For example, the at least one image parameter may include at least one of the following: view position; anatomical features; orientation of the anatomical features; patient identification value; examination identification value; date; and image sensor identification value. Then, when selecting a reference image based on this image parameter, a reference image from the same patient and / or acquired through the same examination procedure or by the same image sensor as the medical image can be selected. The selected reference image may also be an image taken on the same day as the medical image.

[0061] Image recognition models can be implemented using any well-known image processing technique. For example, an image recognition model can be configured to identify and extract information from metadata tags associated with medical images. Alternatively, an image recognition model can include a machine learning model trained to identify imaging parameters of medical images.

[0062] In some embodiments of the invention, sub-steps 122 and 124 may not be present; that is, in some embodiments, selecting a suitable reference image may not require using an image recognition model to process the medical image to determine the image parameters. For example, the medical image may be provided with a label identifying a suitable reference image (e.g., one that has been identified as a suitable reference image by a medical professional), which will be used when analyzing the image to select one from multiple different possible reference images. The key here is that only the information associated with the medical image to be evaluated is needed to select one reference image from multiple different possible reference images. Furthermore, in alternative embodiments, step 150 may also be omitted. Some embodiments may only produce an overall image quality value.

[0063] The overall image quality value determined in method 100 includes at least one of an absolute image quality value and a comparative image quality value. In other embodiments, the overall image quality value may describe the average of the absolute image quality value and the comparative image quality value. Alternatively, the overall image quality value may be derived from the absolute image quality value and the comparative image quality value using other formulas and calculation methods, depending on the relative reliability of these two values. For example, if the comparative image quality value is considered more accurate than the absolute image quality value, the comparative image quality value of the image may be given a greater weight than the absolute image quality value when calculating the overall image quality value.

[0064] Generating a suitable training dataset for a first machine learning model can be challenging. Manually labeling training images using absolute quality scores is often highly subjective and can introduce significant bias into image quality judgments when the model is trained on such labeled datasets. Some embodiments of the present invention attempt to overcome this problem by using pairwise image quality comparisons to compute image quality labels for the training dataset. This can produce better and more consistent image quality labels. Further details of this labeling process will be referenced below. Figure 2 and Figure 3 Let's have a discussion.

[0065] Figure 2This is a simplified diagram of the user interface used during the training data integration and annotation process. First, a training dataset containing suitable medical images is obtained. The dataset is then annotated through the following process: A training image pair, including a first training image 210 and a second training image 220, is displayed to the trained individual acting as the annotator via user interface 200. Then, three annotation results 230, 240, and 250 are presented to the user. Result 230 indicates that the image quality of the first training image 210 is higher than that of the second training image 220. Result 240 indicates that the pair of images is annotated as having the same image quality. Result 250 indicates that the image quality of the second training image 220 is higher. In this example, the user interface is a touch-sensitive screen, so the annotator can then select the annotation result (230, 240, or 250) that they deem most suitable for the presented pair of training images via user interface 200.

[0066] Alternative embodiments may utilize a different user interface configuration than the touch-sensitive user interface. For example, the user interface may include an input interface allowing the user to select annotation results, and an output interface for displaying the pair of training images. For instance, the output interface could be a computer monitor, and the input interface could be a keyboard. Once the annotator selects a annotation result, that result is stored alongside the training database, and another pair of training images is presented to the user for annotation in a similar manner. In this way, a database of annotated pairs of training images can be constructed, where each pair of training images is labeled with a pairwise quality comparison value describing which of the two images (if any) is considered to be of higher quality.

[0067] Annotation networks can be built using this pairwise labeling. Figure 3 A simplified example 300 of this network is depicted. Each circle (or node) in the network represents an image from the labeled training dataset. For example, network 300 depicts images 301, 302, 303, 304, 305, 306, and 307. Arrows (or edges) in the network represent pairwise labels comparing image quality. The arrows start at the image considered to be of higher quality and point to the image considered to be of lower quality. For example, in labeled network 300, image 301 can be seen as being labeled as being of higher quality than images 302 and 303.

[0068] like Figure 3The labeled network shown can then be used to calculate an absolute image quality score for each image in the training dataset. This can be achieved using an Elo-based transformation system. This involves simulating multiple “matches” between images in the training database. Each connection (or arrow) in the labeled network determines the outcome of a match between two associated images, with the image considered to be of higher quality being said to have “won” the game. For each match played, the rating of the winning image increases, and the rating of the losing image decreases. The magnitude by which the ratings of two images are adjusted based on the match outcome is related to the image’s current rating. Once a sufficient number of matches have been simulated, the rating values ​​of the image set tend to stabilize. For example, approximately 200 rounds of iterative matches can be simulated until the average rating change caused by each match is approximately 0.1%. Through this process, each image acquires a quality rating. A higher rating indicates higher image quality, and a lower rating indicates lower image quality.

[0069] Elo-based rating systems naturally follow a Gaussian distribution, with very low and very high ratings being relatively dispersed. To better group images with specific quality levels, a flattening function can be used to flatten the distribution of quality ratings. This flattening process may result in images with the lowest ratings (the bottom 10%) being mapped to a common absolute image quality score. Similarly, images with the highest ratings (the top 10%) can be mapped to a common absolute image quality value. A flattened distribution may be easier for machine learning models to learn. In this way, an absolute image quality value can be obtained for each image in the training database.

[0070] It has been found that using pairwise quality annotations to generate absolute quality values ​​for a set of training images in this manner can significantly improve inter-observer variability among different annotators. Traditional annotation methods require annotators to assign quality values ​​to images in the training dataset (e.g., they might be asked to assign a value of 1-low quality, 2-medium quality, or 3-high quality to each image), but this approach leads to a near-complete lack of consensus among experts. Therefore, using this annotation method to annotate training databases for training machine learning models is unsuitable, as the resulting models may lead to highly subjective image quality ratings. In contrast, the pairwise comparison-based annotation strategy proposed above demonstrates a significant improvement in inter-observer variability among annotators.

[0071] It should be understood that although in the above example, each image yields a single quality value describing the quality of the entire imaging field of view, in alternative embodiments, a similar process can be used to obtain more detailed image quality values ​​associated with each feature present in the image. For example, the training dataset may contain only images of specific anatomical features (e.g., the heart). Two heart training images are then presented to the annotator, who is asked to compare the image quality of each heart feature present in the training images. For example, the annotator may be asked to compare the image quality of the left ventricle, mitral valve, tricuspid valve, aortic valve, pulmonary valve, shortening phenomenon, papillary muscles, and right ventricle. In this case, the absolute quality value of each image obtained through the above process will include multiple quality values ​​describing the corresponding image quality of multiple different anatomical features present in the medical image.

[0072] The annotated dataset of medical images generated in this manner can be used to train the first machine learning model used in the method of the present invention. Specifically, the first machine learning model can be trained using a deep learning algorithm configured to receive an array of training inputs and corresponding known outputs, wherein the training input of the first machine learning model is a first training image, and the corresponding known output is the absolute image quality value of the first training image. The absolute quality value can be calculated based on at least one comparative training image quality value associated with the first training image and describing the quality of the first training image compared to at least one other training image. The absolute image quality value can be calculated based on the at least one comparative training image using an Elo-based transformation system as described above.

[0073] Now for reference Figure 4 The figure depicts a simplified block diagram of a medical imaging system 400 for evaluating the quality of medical images according to the proposed embodiment.

[0074] The medical imaging system 400 includes an image sensor 410, a processing unit 420, a memory assembly 430, and an output interface 440. The image sensor 410 is configured to acquire images of a subject appropriately positioned in front of the medical imaging system 400. This image can then be presented to the operator of the medical imaging system via the output interface 440. The image sensor 410 can be any suitable image sensor for medical imaging. For example, the image sensor 410 can include at least one of the following: an X-ray detector, an MRI scanner, an ultrasound detector, a CT scanner, a PET scanner, and an infrared detector.

[0075] Processing device 420 is configured to evaluate the quality of an image acquired by image sensor 410. It achieves this by: processing the acquired image using a first machine learning model to determine an absolute image quality value; analyzing the image to select a reference image from a plurality of different possible reference images; processing the acquired image and the selected reference image using a second machine learning model to determine a comparative image quality value describing the quality of the acquired image relative to the selected reference image; and determining an overall image quality value of the image acquired by the image sensor based on the absolute image quality value and the comparative image quality value. The plurality of different possible reference images may be stored in a memory component 430 associated with processing device 420. The acquired image may also be stored in memory component 430 and may be used as a possible reference image in subsequent examinations using the same medical imaging system.

[0076] Processing device 420 and memory component 430 can be implemented in various ways, including software and / or hardware, to perform a variety of required functions. Processing device 420 may typically employ one or more microprocessors that can be programmed using software (e.g., microcode) to perform the required functions. Processing device 420 can be implemented as a combination of dedicated hardware for performing some functions and one or more programmable microprocessors and associated circuitry for performing other functions. Examples of circuitry that can 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).

[0077] The memory component 430 associated with the processing device 420 may include one or more storage media, such as volatile and non-volatile computer memories like RAM, PROM, EPROM, and EEPROM. The storage media may be encoded with one or more programs that, when run on one or more processors and / or controllers, perform the required functions. Various storage media may be fixed within the processor or controller, or may be removable, allowing one or more programs stored thereon to be loaded into the processor. The output interface 440 is also configured to output an indication of the determined overall image quality value to the operator of the medical imaging system. Examples of this configuration include... Figure 5 As shown.

[0078] Figure 5 An example output interface 500 of a system for evaluating the quality of medical images according to the proposed embodiment is described.

[0079] Output interface 500 displays medical image 510 and associated quality scores 1-3 (520, 530, and 540) and a comparison rating 550. Quality scores 520, 530, and 540, along with the comparison rating 550, can collectively constitute the overall image quality value of this embodiment of the invention. In this exemplary embodiment, quality scores 1-3 can be absolute image quality scores, each associated with a different anatomical feature present in image 510. For example, using the method proposed in this invention, it can be determined that medical image 510 has high image quality on a first feature corresponding to quality score 1, and low image quality on second and third features corresponding to quality scores 2 and 3. This can be presented to the operator as a quantitative score (i.e., a 10-point scale, where 10 indicates that a feature has been imaged with high quality) or as a qualitative assessment (i.e., high quality or low quality). Alternatively, quality scores 1-3 can be presented in the form of an applicability assessment, i.e., they can represent a simple assessment of whether the image quality is “sufficient” or “insufficient” for diagnostic purposes. Quality scores 1-3 can be displayed as numbers, text, or via color indicators. Although three quality scores are shown in this example, in other embodiments, any number of quality scores associated with different features present in the medical image 510 can be displayed to the user on the output interface 500.

[0080] The comparison rating can be presented to the user as an indication of the comparative image quality value determined for medical image 510 using any of the methods proposed in the embodiments. For example, the comparison rating can indicate to the operator whether higher quality images have been obtained in the past. For example, the comparison rating can indicate to the operator in words that previously acquired images could have better image quality for features 2 and 3 depicted in medical image 510. In this way, information can be presented to the operator prompting them to try to obtain more medical images with higher image quality for the same patient on features 2 and 3.

[0081] Figure 6 An example of a computer 600 that may employ one or more portions of an embodiment is shown. The various operations discussed above can utilize the capabilities of computer 600. In this regard, it should be understood that system functional blocks may run on a single computer or may be distributed across multiple computers and locations (e.g., via an internet connection).

[0082] Computer 600 includes, but is not limited to, PCs, workstations, laptops, PDAs, handheld devices, servers, storage devices, etc. Typically, in terms of hardware architecture, computer 600 may include one or more processors 610, memory 620, and one or more I / O devices 630 communicatively coupled via a local interface (not shown). As known in the art, the local interface may be, for example, but not limited to, one or more buses or other wired or wireless connections. The local interface may have additional elements, such as controllers, buffers (caches), drivers, repeaters, and receivers, to enable communication. Furthermore, the local interface may include address, control, and / or data connections to enable appropriate communication between the aforementioned components.

[0083] Processor 610 is a hardware device for executing software that can be stored in memory 620. Processor 610 can be virtually any custom or commercially available processor, central processing unit (CPU), digital signal processor (DSP), or auxiliary processor among multiple processors associated with computer 600, and processor 610 can be a semiconductor-based microprocessor (in the form of a microchip) or microprocessor.

[0084] Memory 620 may include any or a combination of volatile memory elements (e.g., random access memory (RAM), such as dynamic random access memory (DRAM), static random access memory (SRAM), etc.) and non-volatile memory elements (e.g., ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic tape, optical disc read-only memory (CD-ROM), magnetic disk, floppy disk, cassette tape, etc.). Furthermore, memory 620 may contain electronic, magnetic, optical, and / or other types of storage media. Note that memory 620 may have a distributed architecture, where various components are geographically separated but still accessible by processor 610.

[0085] The software in memory 620 may include one or more independent programs, each containing an ordered list of executable instructions for implementing logical functions. According to an exemplary embodiment, the software in memory 620 includes a suitable operating system (O / S) 650, a compiler 660, source code 670, and one or more application programs 680. As shown, application program 680 includes numerous functional components for implementing the features and operations of the exemplary embodiment. Application program 680 of computer 600 may represent various application programs, computing units, logical units, functional units, processes, operations, virtual entities, and / or modules according to the exemplary embodiment, but this does not imply limitation on application program 680.

[0086] Operating system 650 controls the execution of other computer programs and provides scheduling, input / output control, file and data management, memory management, communication control, and related services. The inventors anticipate that application program 680 for implementing the exemplary embodiments can be applied to all commercially available operating systems.

[0087] Application 680 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 typically translated by tools such as a compiler (e.g., compiler 660), an assembler, an interpreter, etc. (these tools may or may not be contained in memory 620) to operate appropriately in conjunction with operating system 650. Furthermore, application 680 can be written in an object-oriented programming language with data and method classes, or in a procedural programming language with routines, subroutines, and / or functions, such as, but not limited to, C, C++, C#, Pascal, Python, BASIC, API calls, HTML, XHTML, XML, ASP scripts, JavaScript, FORTRAN, COBOL, Perl, Java, ADA, .NET, etc.

[0088] I / O device 630 may include input devices, such as, but not limited to, a mouse, keyboard, scanner, microphone, camera, etc. Furthermore, I / O device 630 may also include output devices, such as, but not limited to, a printer, monitor, etc. Finally, I / O device 630 may also include devices that can both input and output, such as, but not limited to, a NIC or modem (for accessing remote devices, other files, devices, systems, or networks), radio frequency (RF) or other transceivers, telephone interfaces, bridges, routers, etc. I / O device 630 also includes components for communication over various networks, such as the Internet or intranets.

[0089] If the computer 600 is a PC, workstation, intelligent device, etc., the software in the memory 620 may also include a Basic Input / Output System (BIOS) (omitted for simplicity). The BIOS is a set of basic software routines that initialize and test the hardware at startup, start the operating system 650, and support data transfer between hardware devices. The BIOS is stored in some type of read-only memory, such as ROM, PROM, EPROM, EEPROM, etc., so that the computer 600 can execute the BIOS at startup.

[0090] When the computer 600 is running, the processor 610 is configured to execute software stored in the memory 620 to transfer data to and receive data from the memory 620, and typically controls the operation of the computer 600 according to the software. The application program 680 and the operating system 650 are read, in whole or in part, by the processor 610, possibly buffered within the processor 610, and then executed.

[0091] When application 680 is implemented as software, it should be noted that application 680 can be stored on virtually any computer-readable medium for use by or in connection with any computer-related system or method. In the context of this document, a computer-readable medium can be an electronic, magnetic, optical, or other physical device or apparatus that can contain or store computer programs for use by or in connection with a computer-related system or method.

[0092] Application 680 may be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device (e.g., a computer-based system, a processor-containing system, or other system that can fetch and execute instructions from and to an instruction execution system, apparatus, or device). In the context of this document, "computer-readable medium" can be any means that can store, transmit, propagate, or transfer a program for use by or in connection with an instruction execution system, apparatus, or device. Computer-readable media may be, for example, but not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, devices, or propagation media.

[0093] Figure 1 Methods and Figure 4 The system can be implemented in hardware, software, or a combination of both (e.g., as firmware running on a hardware device). In embodiments where some or all of the implementation is in software, the functional steps shown in the process flowchart can be executed by a suitably programmed physical computing device, such as one or more central processing units (CPUs) or graphics processing units (GPUs). Each process and its respective component steps as shown in the flowchart can be executed by the same or different computing devices. According to an embodiment, a computer-readable storage medium stores a computer program including computer program code that, when run on one or more physical computing devices, is configured to cause the one or more physical computing devices to perform the methods described above.

[0094] Storage media can include volatile and non-volatile computer memories, such as RAM, PROM, EPROM and EEPROM, optical discs (such as CD, DVD, BD), and magnetic storage media (such as hard disks and magnetic tapes). Various storage media can be fixed within a computing device or can be removable, allowing one or more programs stored on them to be loaded into a processor.

[0095] In cases where some or all of the embodiments are implemented in hardware. Figure 4 The blocks shown in the block diagram can be individual physical components, logical subdivisions of a single physical component, or all integrated into a single physical component. The functionality of one block shown in the figures can be divided among multiple components in the implementation, or the functionality of multiple blocks shown in the figures can be combined into a single component in the implementation. Hardware components suitable for embodiments of the present invention include, but are not limited to, conventional microprocessors, application-specific integrated circuits (ASICs), and field-programmable gate arrays (FPGAs). One or more blocks can be implemented as a combination of dedicated hardware for performing some functions and one or more programmable microprocessors and associated circuitry for performing other functions.

[0096] By studying the accompanying drawings, disclosure, and claims, those skilled in the art can understand and implement variations of the disclosed embodiments when carrying out the claimed invention. In the claims, the word "comprising" does not exclude other elements or steps, and the quantifiers "a" or "an" do not exclude a plurality. A single processor or other unit can perform the functions of several items recited in the claims. The fact that certain measures are recited in mutually different dependent claims does not mean that combinations of these measures cannot be used advantageously. If a computer program has been discussed above, it can be stored / distributed on a suitable medium, such as an optical storage medium or solid-state medium provided with or as part of other hardware, but it can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems. If the term "suitable" is used in the claims or description, it should be noted that the term "suitable" is intended to be equivalent to the term "configured as." Any reference numerals in the claims should not be construed as limiting the scope.

[0097] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in a flowchart or block diagram may represent a module, segment, or portion of instructions, including one or more executable instructions for implementing a specified logical function(s). In some alternative embodiments, the functions marked in the boxes may occur in a non-linear order. For example, two boxes shown consecutively may actually be executed substantially simultaneously, or these boxes may sometimes be executed in reverse order depending on the functions involved. It should also be noted that each box in the block diagrams and / or flowcharts, and combinations of boxes in the block diagrams and / or flowcharts, may be implemented by a system based on dedicated hardware that performs the specified function or action or performs a combination of dedicated hardware and computer instructions, illustrating the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each box in a flowchart or block diagram may represent a module, segment, or portion of instructions, including one or more executable instructions for implementing a specified logical function(s). In some alternative embodiments, the functions marked in the boxes may occur in a non-linear order. For example, two boxes shown consecutively may actually be executed substantially simultaneously, or these boxes may sometimes be executed in reverse order depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart illustrations, and combinations of boxes in the block diagram and / or flowchart illustrations, may be implemented by a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions.

Claims

1. A computer-implemented method (100) for evaluating the quality of medical images, the method comprising: The medical image is processed (110) using a first machine learning model to determine the absolute image quality value of the medical image; Analyze the medical image described in (120) to select a reference image from a plurality of different possible reference images; The medical image and the selected reference image are processed using a second machine learning model (130) to determine a comparative image quality value that describes the quality of the medical image relative to the selected reference image; and The overall image quality value of the medical image is determined (140) based on the absolute image quality value and the comparative image quality value.

2. The method according to claim 1, wherein, Analyzing the medical images to select a reference image also includes: The medical image is processed using an image recognition model (122) to determine at least one image parameter of the medical image; Based on at least one determined image parameter, a reference image is selected (124) from a plurality of different possible reference images.

3. The method according to claim 2, wherein, The at least one image parameter includes at least one of the following: view position; anatomical features; Orientation of anatomical features; patient identification value; examination identification value; date; and image sensor identification value.

4. The method according to any one of claims 1-3, wherein: The first machine learning model is trained using a deep learning algorithm, which is configured to receive an array of training inputs and corresponding known outputs. Wherein, one training input of the first machine learning model is the first training image (210) in a pair of training images (210, 220), and the corresponding known output is an absolute image quality value, and The absolute image quality value is calculated based on the comparative training image quality value associated with the pair of training images, wherein the comparative training image quality value describes the quality of the first training image (210) relative to the second training image (220) in the pair of training images (210, 220) based on the annotations (230, 240, 250) of the trained individuals.

5. The method according to claim 4, wherein, Calculating the absolute image quality value of the first training image (210) based on the comparative training image quality value associated with the pair of training images (210, 220) includes using an Elo-based transformation system.

6. The method according to any one of claims 1-5, wherein, The absolute image quality value includes multiple values ​​that describe the corresponding image quality of multiple different anatomical features present in the medical image.

7. The method according to any one of claims 1-6, wherein, The selected reference image has a substantially identical view orientation to the medical image to be evaluated.

8. The method according to any one of claims 1-7, wherein, The selected reference image depicts the same anatomical features as the medical image to be evaluated.

9. The method according to claim 8, wherein, The comparative image quality values ​​include multiple values ​​that describe the relative image quality of each anatomical feature present in the medical image to be evaluated compared to the image quality of the same anatomical feature depicted in the reference image.

10. The method according to any one of claims 1-9, further comprising: The suitability value of the medical image (150) is determined by comparing the determined overall quality value with a first threshold.

11. The method according to any one of claims 1-10, wherein, The overall quality value includes at least one of the following: the absolute image quality value; and the comparative image quality value.

12. A computer program including code modules, wherein when the program is run on a processing system, the code modules are used to implement the method (100) of any of the preceding claims.

13. A processing apparatus (420) configured to: The medical image is processed using a first machine learning model (110) to determine the absolute image quality value of the medical image; Analyze the medical image described in (120) to select a reference image from a plurality of different possible reference images; The medical image and the selected reference image are processed using a second machine learning model (130) to determine a comparative image quality value that describes the quality of the medical image relative to the selected reference image; and The overall image quality value of the medical image is determined (140) based on the absolute image quality value and the comparative image quality value.

14. A medical imaging system (400) suitable for evaluating the quality of medical images, the system comprising the processing device (420) according to claim 13.

15. The system of claim 14, further comprising at least one of the following: Image sensor (410); and An output interface (440) is configured to output an indication of the overall image quality of the determined medical image.

Images