Global Medical Imaging Patent Analysis Report: The Midfield Battle of the GPS Triad

Publication (Announcement) No.: KR1020130080312A

Claim 1: A method for measuring biometric features of a subject, the method comprising: receiving an image of the subject; simulating the subject to identify at least a portion of the subject; and measuring the biometric features of the subject based on a result of the simulation.

Technical Issue: Biometric parameters such as crown-rump length (CRL), nuchal translucency (NT), and intracranial translucency (IT) are measured and reported individually. However, the relative discrepancy between NT or IT and CRL—specifically, values derived from at least two biometric parameters—is utilized to assess fetal status. Therefore, it is necessary to automatically provide users with values calculated based on a comprehensive analysis of biometric parameters like CRL, NT, and IT, along with diagnostic results regarding fetal status derived from these calculated values, thereby enabling early diagnosis and determination of fetal condition.

Technical Effect: A method and apparatus for measuring biometric features of a subject are provided, enabling automatic measurement of the subject's biometric features using ultrasound images of the subject.

Brief Analysis: This patent enables automated biometric measurement of a subject by receiving image simulations of the subject and processing the simulation results. The patent is currently in force and demonstrates strong technical stability. It has been cited 13 times worldwide, including its patent family members, indicating its technological advancement. With 22 claims and filings in 10 countries and regions—including the United States, China, the European Patent Office, South Korea, and India—the patent offers robust protection scope, reflecting the company’s significant emphasis on this intellectual property asset.

Publication (Announcement) Number: EP2733669B1

Claim 1: An apparatus, comprising: an analysis unit (110) configured to detect lesion regions from medical images and generate a set of candidate lesion regions relative to the detected lesion regions; and an interface unit (130) configured to arrange and display, in a first area (20) of an interface, the group of candidate lesion regions (21) with brief analytical information based on a priority determined according to predetermined criteria including class information about each candidate lesion region, and to display detailed analytical information about one of the selected candidate lesion regions in a second area (30) of the interface if a user selects one of the candidate lesion regions displayed in the first area (20), wherein the class information is used to indicate the degree of benignity or malignancy of the candidate lesion regions, wherein candidate lesion regions with higher priority in the group of candidate lesion regions are displayed in the first area (20) with at least one aspect relative to the group, the group comprising: being in a more central position; having a larger size; and having a color that is more noticeable than candidate lesion regions with lower priority in the group of candidate lesion regions.

Technical Problem: Computer-aided diagnosis (CAD) is a technology used in medicine to assist physicians in interpreting medical images by detecting suspected abnormal regions within the images and analyzing these regions to provide a preliminary diagnosis of lesions. In existing technologies, to analyze the diagnostic results of a CAD system, users must evaluate various images of the abnormal regions from different perspectives. However, such evaluation is constrained by temporal and spatial limitations, making it difficult to analyze and modify the diagnostic results provided by the CAD system through the user interface offered by such systems.

Technical Effect: A diagnostic method and apparatus are provided, enabling the diagnosis and analysis of lesion regions.

Brief Analysis: This patent detects lesion regions in medical images, generates a set of candidate lesion regions along with their respective priorities relative to the detected lesion areas, and displays the group of candidate lesion regions with concise analytical information to achieve diagnostic analysis of pathological regions. The patent is currently in force and demonstrates good technical stability. It has been cited 19 times globally, including its patent family, indicating strong technological advancement. The patent has been filed in five countries and regions—the United States, China, the European Patent Office, Japan, and South Korea—providing robust scope of protection.

Publication (Announcement) No.: KR1020040059312A

Claim 1: A method for extracting a region of interest from a tongue image, the method comprising: constructing a database storing template images, wherein each template image corresponds to personal information and displays a region of interest; segmenting a tongue region from a tongue image obtained from a subject whose health status is to be determined; matching the tongue region with the template images stored in the database; and extracting the region of interest from the matched template image.

Technical Issue: In the prior art, devices for detecting the early stages of disease onset and progression based on tongue conditions—such as tongue coating, color, or appearance—are prohibitively expensive and are predominantly available only in hospital settings. Similarly, conventional devices are unsuitable for personal health monitoring by individuals.

Technical Effect: Achieves health monitoring by extracting a region of interest from tongue images and analyzing at least one characteristic factor from the extracted region of interest.

Brief Analysis: This patent enables health monitoring by extracting regions of interest from tongue images and analyzing at least one characteristic factor from these extracted regions. The patent is currently in force and demonstrates strong technical stability. It, along with its patent family, has been cited 87 times worldwide, indicating a high level of technological advancement. The patent has been filed in five countries and regions—the United States, China, the European Patent Office, Japan, and South Korea—providing robust scope of protection.

Publication (Announcement) Number: US9799115B2

Claim 1: A method for automatically registering landmarks in a three-dimensional (3D) medical image of an object, the method comprising: obtaining the 3D image; determining a set of search points based on a statistical atlas attached to a bounding box corresponding to a portion of the 3D image, wherein the statistical atlas includes information indicating a statistical distribution of previously identified landmarks in the object; extracting features from the determined set of search points; forming a set of landmark candidates based on the extracted features; filtering the candidates and outputting remaining candidates from among the candidates based on the filtering; and outputting a final position of the landmark based on one of the remaining candidates.

Technical Challenges: Most registration algorithms are designed to work exclusively with two-dimensional medical images, limiting their applicability to specific domains. Some methods can only process high-resolution, high-quality medical images, which require prolonged acquisition times. In practice, feature extraction algorithms in these approaches are typically employed to address classification tasks. Generally, these feature extraction algorithms are manually tuned based on empirical knowledge. In certain scenarios, generic feature extraction models are trained without accounting for the specific characteristics of the data being used, thereby failing to fully leverage the potential of machine learning methods. Furthermore, most approaches construct only a single level of features rather than establishing a multi-level hierarchical feature representation. In some methods, candidate landmarks are filtered using corresponding thresholds; however, these thresholds are also manually adjusted as rule-based parameters.

Technical Effect: Provided is an improved method for automatically registering landmarks in three-dimensional (3D) medical images, which allows for the detection of the positions of unique points within the constraints specific to each type of such points, as well as the positions of multiple points for each of these types.

Brief Analysis: This patent is currently in force and exhibits strong technical stability. It and its patent family have been cited five times globally, indicating a high level of technological advancement. The patent has been filed in five jurisdictions—the United States, the World Intellectual Property Organization (WIPO), the European Patent Office (EPO), Russia, and South Korea—providing robust scope of protection.

Publication (Announcement) Number: FR2804516A1

Claim 1: An ultrasound imaging method for forming an image of an object using signals reflected from the object after transmitting ultrasound pulses to the object, the method comprising the steps of: (a) transmitting a first set of ultrasound pulses to the object by applying voltages to one or more transducers (1) according to a first code of a pair of modified first Golay codes; (b) performing pulse compression on a first set of reflected signals of the first set of ultrasound pulses reflected from the object; (c) transmitting a second set of ultrasound pulses toward the object to the one or more transducers (1) by means of applied voltages according to a second code of the pair of modified Golay codes; (d) performing pulse compression on a second set of reflected signals of the second set of ultrasound pulses reflected from the object; (e) summing the compressed pulses and signals of the first and second sets of reflected signals; (f) generating a signal after receiving focusing using the summed signal to form an image of the object; and (g) displaying the image.

Technical Issue: Due to the sidelobe suppression characteristics of Golay codes, efforts have been made to apply them in long-pulse ultrasound imaging systems. However, one undesirable frequency characteristic of Golay codes is their broader spectrum compared to conventional ultrasound transducers. This results in some power loss of the Golay code on the ultrasound transducer, preventing the system's signal-to-noise ratio (SNR) from reaching the desired level.

Technical Effect: To provide codes whose frequency characteristics match those of an ultrasonic transducer, and an ultrasonic imaging method based on pulse compression technology using these codes.

Brief Analysis: This patent and its patent family have been cited 57 times worldwide, indicating strong technological advancement. The patent has been filed in eight countries and regions, including the United States, Germany, France, the European Patent Office, Japan, and South Korea, demonstrating broad scope of protection and reflecting the company’s high level of attention to this patent.

Publication (Announcement) No.: CN110337673A

Technical Branch: Image Processing Technology Field, Reconstruction Technology Direction

Technical Issue: Due to various reasons, a portion of the projection data for a given scan may be corrupted or missing (relative to an ideal or mathematically complete projection dataset), which can consequently lead to image artifacts. Traditional interpolation, extrapolation, or iterative estimation techniques are not always capable of adequately addressing such instances of missing or incomplete data, and in some cases, computational speed may be slow.

Technical Solution: A method for estimating missing data used in tomographic reconstruction, the method comprising: acquiring or accessing a set of scan data, wherein the set of scan data has one or more regions of incomplete or unsuitable data; processing the set of scan data using one or more trained neural networks; generating, using the one or more trained neural networks, an estimated dataset for each region of incomplete or unsuitable data, wherein the set of scan data together with the estimated dataset corresponds to a corrected set of scan data; and performing tomographic reconstruction on the corrected set of scan data to generate one or more reconstructed images.

Technical Effects: The method involves using deep learning techniques that can be implemented with a trained neural network (50) to estimate various types of missing projection data or other non-reconstructed data. Similarly, as opposed to estimating missing projection data, the method may also be used to replace or correct corrupted or erroneous projection data.

Brief Analysis: This patent is currently under examination and demonstrates strong technical stability. It contains 30 claims and has six INPADOC family members, with patent applications filed at the European Patent Office, in Japan, South Korea, the United States, and through the World Intellectual Property Organization. The broad scope of patent protection reflects the company’s significant emphasis on this patent.

Publication (Announcement) No.: CN110807737A

Technical Branch: Image Processing Technology Field, Reconstruction Technology Direction

Technical Issue: Computed tomography (CT) images may be affected by image noise due to the quantum nature of X-rays and detector readout noise. Such images are limited in spatial resolution by several factors, including finite focal spot size, detector element size, and the reconstruction process. Image noise increases when the radiation dose is increased or when spatial resolution is improved. Therefore, it is desirable to minimize image noise and maximize spatial resolution for a given radiation dose. It is also important that the reconstruction process preserves low-contrast details and is computationally efficient.

Technical Solution: A method for reconstructing an image, comprising: receiving a sinogram input; generating one or more intermediate sinograms based on the sinogram input or based on one or more intermediate images generated from the sinogram input; iteratively processing the intermediate sinograms, wherein the iterative processing includes at least performing an iterative data fitting operation to compare an output of an iterative loop with an input to the iterative loop; in addition to the data fitting operation, performing a denoising operation, wherein the denoising operation is performed using a trained artificial neural network; and after completing the iterative processing, reconstructing a final image and outputting the final image for viewing, storage, or further processing.

Technical Effects: This method combines the simplicity of direct reconstruction with the denoising capabilities of deep learning, while also leveraging the statistical benefits of data fidelity terms. Furthermore, in various specific implementations, the present disclosure can be readily integrated into different CT platforms, for example, by incorporating a first-pass reconstruction step.

Brief Analysis: This patent is currently under examination and demonstrates good technical stability. It contains 10 claims and has been filed in four countries and regions, including the United States, China, the European Patent Office, and Japan. The broad scope of patent protection reflects the company’s high level of attention to this patent.

Publication (Announcement) No.: US20130060121A1

Technical Branch: Image Processing Technology Field, Recognition Technology Direction

Technical Issue: Currently, imaging techniques such as computed tomography (CT), magnetic resonance (MR) imaging, and X-ray are used for the diagnosis of arthritis. However, modalities like X-ray rely on 2D projections of anatomical structures and cannot generate accurate images of the underlying 3D anatomy. Moreover, other imaging methods such as CT and MR are relatively expensive and contraindicated for certain patient groups. Therefore, there is a need to design and develop specialized methods and systems that provide faster and more accurate pathological diagnosis and assessment of treatment response for conditions such as musculoskeletal pathologies. In particular, there is a desire to develop systems and methods for the detection, quantification, and/or tracking of pathologies that allow for easier use, shorter learning curves, faster examination times, and reduced operator dependence.

Technical Solution: An automated method for detecting disease status, the method comprising: identifying bone surfaces in one or more image datasets, wherein the one or more datasets correspond to regions of interest in an object of interest; segmenting joint capsule regions corresponding to the one or more image datasets based on the identified bone surfaces; and analyzing the segmented joint capsule regions to identify the disease status.

Technical Effects: The described methods and systems facilitate early diagnosis, quantification (scoring), and enhanced longitudinal tracking of pathology, while reducing operator dependency in pathological assessment. Furthermore, an objective method for evaluating pathology is proposed, thereby improving the efficiency of pathological diagnosis.

Brief Analysis: The patent is currently in force and exhibits strong technical stability. It and its patent family have been cited 18 times globally, indicating a high level of technological advancement. With 21 claims and filings in four jurisdictions—the United States, China, Japan, and Germany—the patent offers robust protection scope, reflecting the company’s significant emphasis on this asset.

Publication (Announcement) Number: WO03069553A2

Technical Branch: Image Processing Technology Field, Recognition Technology Direction

Technical Issue: MR imaging equipment is used to generate images of known tumors as a "training set." The same equipment is also employed to provide a set of test samples from the scanned body region for detecting secondary tumors. Similarity data are then provided to indicate the degree of similarity between the test samples and training samples, preferably by determining the Euclidean distance between each member of the training set and the test set. A disadvantage of this method is that the training set is defined based on known tumors in the images; in most cases, this training set is inherently heterogeneous, causing adjacent pixels in tumor images to exhibit markedly different behaviors. Another drawback is the lack of appropriate validation for the constructed classes and the absence of physiological information in the processed datasets.

Technical Solution: A method for analyzing image data, comprising: a) generating data in image space by acquiring multi-channel data in MR tomography of a human or non-human animal body, wherein at least one subset of the channels describes the dynamic behavior of an MR contrast agent previously administered to said body; b) defining at least one region of interest (ROI) in the image space; c) converting the image data generated in step a) or data corresponding to the ROI into score map data using multivariate image analysis, thereby generating data in score map space; d) determining a region of interest in the score map space corresponding to the ROI; e) selecting relevant data points associated with the ROI; and f) mapping the data points selected in step e) back to the image space, thereby identifying image data having properties similar to those of the ROI.

Technical Effect: The present invention relates to a method for analyzing image data obtained in magnetic resonance tomography, and to the use of said method in the identification of pathological tissue, preferably tumor tissue.

Brief Analysis: This patent and its patent family have been cited 27 times worldwide, indicating strong technological advancement. The patent contains 12 claims and has been filed in five countries and regions, including the United States, Australia, the European Patent Office, the World Intellectual Property Organization, and Norway, demonstrating broad scope of protection. This reflects the company’s high level of attention to this patent.

Publication (Announcement) No.: US20200012884A1

Technical Branch: Intelligent Decision-Making Technology Field, Disease Prediction Technology Direction

Technical Problem: Artificial intelligence (AI) can be used for the classification and/or analysis of digital images. For example, AI can be employed for image recognition. In certain technical applications, AI can be utilized to enhance imaging analysis. In one example, a region-of-interest-based deep neural network can be used to localize features within digital images. However, it is often difficult to achieve the accuracy and/or efficiency of classifying and/or analyzing digital images using conventional manual techniques. Furthermore, traditional manual techniques for the classification and/or analysis of digital images typically require labor-intensive processes such as pixel annotation and voxel-level annotation. Therefore, there is room for improvement in traditional manual techniques for the classification and/or analysis of digital images.

Technical Solution: A machine learning system, comprising: a memory storing computer-executable components; and a processor executing the computer-executable components stored in the memory, wherein the computer-executable components comprise: a training component that trains a convolutional neural network based on training data and a plurality of images; wherein the training data is associated with multiple patients from at least one imaging device, and wherein the plurality of images is associated with a plurality of masks from multiple objects; a first loss function component that generates a first loss function based on the plurality of masks; a second loss function component that generates a second loss function based on a plurality of image-level labels associated with the plurality of images; a third loss function component that generates a third loss function based on the first loss function and the second loss function, wherein the third loss function is iteratively backpropagated to tune parameters of the convolutional neural network; and a classification component that predicts a classification label for an input image based on the convolutional neural network.

Technical Effects: By providing richer annotation information (e.g., masks), classification accuracy can be improved, and the convolutional neural network can also output refined localization maps. This can be achieved using the same underlying prediction model for both tasks. Furthermore, the system can apply classification and/or localization to disease detection (e.g., medical condition detection) in medical imaging data (e.g., X-ray images) and/or other digital images.

Brief Analysis: This patent has 20 claims and has been filed in two jurisdictions, namely the United States and the World Intellectual Property Organization (WIPO), offering robust patent protection scope.