Optimized organ image processing using ai

Abstract

A technique to optimize medical image enhancement facilitated by AI / deep learning neural network implementations. In various embodiments, a computer executable component can include: a receiving component that receives a set of "region / volume of interest" images containing multiple organs; and an artificial intelligence deep learning neural network model component that automatically processes and enhances the respective images in a locally adaptive manner such that at each location, the enhanced image is optimized for the organ displayed at that location.

Description

[0001] Related patent applications

[0002] This application claims priority to U.S. Provisional Patent Application No. 63 / 601,479, filed November 21, 2023, entitled "OPTIMIZED ORGAN IMAGE PROCESSING USING AI," and U.S. Non-Provisional Application Serial No. 18 / 951,587, filed November 18, 2024, both entitled "OPTIMIZED ORGAN IMAGE PROCESSING USING AI," the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure relates in general to automated artificial intelligence (AI) solutions for contrast enhancement of organs and body regions, such as computed tomography (CT) images with or without intravenous contrast agents. Background Technology

[0004] When given a medical image, if the region of interest is not detailed or clear enough, medical personnel may find it difficult to understand the image's precise meaning. Contrast in medical images is essential for accurate diagnosis and interpretation. However, problems related to contrast in medical images can exist that may affect the quality and reliability of diagnostic information. Brightness (such as contrast) in medical images plays a crucial role in diagnostic interpretation. Brightness-related problems can affect the visibility and interpretation of structures, potentially impacting diagnostic accuracy. Summary of the Invention

[0005] The following summary is presented to provide a basic understanding of one or more embodiments of the invention. This summary is not intended to identify key or essential elements, nor is it intended to depict any scope of the specific embodiments or any scope of the related claims. Its sole purpose is to present concepts in a simplified form as a prelude to the more detailed description that follows. In one or more embodiments described herein, apparatus, systems, computer-implemented methods, devices, devices, or computer program products that facilitate improved deep learning image processing are described.

[0006] CT images, such as molecular imaging computed tomography (MICT), have been used in many different diagnostic applications due to their ability to enhance the appearance of organs or tumors when scanning across different stages of pharmacokinetic activity. However, the automated raw images provide a general appearance by default, which often requires different presets / reconstructions / scans for brightness and contrast to enhance the organ / region of interest. For example, the lungs and bones may have different preferred visualization settings / reconstructions for optimized visualization. Therefore, in cases where radiologists want to examine different organs or regions, multiple image reconstructions and manipulations between different views are necessary. Clinicians often need to rely on their imagination to generate composite and combined images in their minds.

[0007] To accelerate visualization and optimize CT images of different organs and regions, an automated image enhancement method is needed. The implementation scheme is an AI-based technique for CT image enhancement. This technique can automatically optimize organ contrast for different preferred views while still preserving structure and detail within the image. This innovation provides a seamless optimization solution for CT image diagnosis. Advantageously, the workload and cost of multi-region analysis in 2D / 3D / 4D images are significantly reduced. This innovation can leverage an automated, AI-based solution to reduce the workload of radiologists reviewing images of different organs and / or tumors. This innovation provides a method to reduce the complexity of image manipulation between different views.

[0008] In summary, region-based automatic image contrast enhancement technology provides optimized AI-based automatic CT image enhancement. This innovation is a fully automated solution, thus eliminating the need for prior organ masks to select regions of interest (ROIs) for corresponding brightness and contrast manipulation. Due to end-to-end training using a remapping algorithm, the preserved results are smooth and anatomically accurate, with smooth transitions between regions and preservation of structural details, representing a significant improvement over conventional implementations (e.g., those utilizing common image-to-image conversion algorithms). This alleviates the burden on radiologists / clinicians of mentally fusing different organs from multiple images. The final output of the AI ​​model generates composite images with corresponding desired / optimal region enhancements; therefore, users can avoid tedious image adjustments in their daily diagnostic work. Training clinicians using this method is straightforward, and the innovation is highly intuitive for clinicians learning fundamental concepts and radiological features. Overall, this innovation is relatively easy to train and requires less annotation and training effort compared to conventional methods used to achieve similar results. Compared to conventional tools (e.g., those utilizing complex segmentation models), the results are easier to maintain; the innovation is simple and easy to train and maintain for modification or fine-tuning.

[0009] According to one or more embodiments, a system is provided. The system may include a non-transitory computer-readable storage medium for storing computer-executable components. The system may also include a processor operatively coupled to and executable on the non-transitory computer-readable storage medium, and capable of executing the computer-executable components stored therein. In various embodiments, the computer-executable components may include: a receiving component that receives a set of "regions of interest / volumes" images comprising multiple organs; and an artificial intelligence deep learning neural network model component that automatically processes and enhances the corresponding images in a locally adaptive manner, such that at each location, the enhanced image is optimized for the organ displayed at that location.

[0010] In another embodiment, a computer-implemented method includes: receiving, by a device operatively coupled to a processor, a set of "regions of interest / volumes" images containing multiple organs; and employing an artificial intelligence deep learning neural network model that automatically processes and enhances the images in a locally adaptive manner, such that at each location, the enhanced image is optimized for the organ displayed at that location.

[0011] In yet another embodiment, a computer program product for facilitating automated organ image enhancement includes a non-transitory computer-readable storage having program instructions embodied therein, which are executable by a processor to cause the processor to: receive, via a device operatively coupled to the processor, a set of "regions of interest / volumes" images containing multiple organs; and employ an artificial intelligence deep learning neural network model by the processor to automatically process and enhance the corresponding images in a locally adaptive manner, such that at each location, the enhanced image is optimized for the organ displayed at that location.

[0012] According to one or more implementation schemes, the above system can be implemented as a computer-implemented method or a computer program product. Attached Figure Description

[0013] Figure 1 A block diagram illustrating an example non-limiting system for facilitating enhanced image processing according to one or more embodiments described herein is shown.

[0014] Figure 2 Examples of various images with different brightness and contrast according to one or more embodiments described herein are illustrated.

[0015] Figure 3 Examples of image optimization processes according to one or more embodiments described herein are illustrated.

[0016] Figure 4 Example non-limiting block diagrams illustrating how a deep learning neural network can be trained on a training dataset with reference to a benchmark ground truth are shown according to one or more embodiments described herein.

[0017] Figure 5 Figure 7B illustrates examples of current segmentation-based images and images processed by the present invention according to one or more embodiments described herein.

[0018] Figure 8 A flowchart illustrating an example non-limiting computer-implemented method for facilitating enhanced image processing according to one or more embodiments described herein is provided.

[0019] Figure 9 Examples are illustrated of using text as prompts to guide the image optimization process of feature decoding according to one or more embodiments described herein.

[0020] Figure 10 A block diagram illustrating an example non-limiting operating environment in which one or more embodiments described herein may be facilitated.

[0021] Figure 11 Example networking environments are illustrated that can operate to perform the various specific implementations described herein. Detailed Implementation

[0022] The following specific embodiments are merely illustrative and are not intended to limit the implementation or application / use of the embodiments. Furthermore, they are not intended to be construed as being bound by any express or implied information presented in the foregoing "Background Art" or "Summary of the Invention" or "Detailed Description" sections.

[0023] One or more embodiments will now be described with reference to the accompanying drawings, wherein the same reference numerals are used to denote the same elements throughout. In the following description, numerous specific details are set forth for purposes of explanation in order to provide a more thorough understanding of the one or more embodiments. However, it will be apparent, in various cases, that one or more embodiments may be practiced without these specific details.

[0024] Medical imaging plays a vital role in modern healthcare and is an integral part of the diagnosis, treatment, and monitoring of a wide range of medical conditions. Here are some reasons why imaging is so important in medicine: For diagnosis, medical imaging techniques such as X-rays, CT scans, MRI, ultrasound, and nuclear medicine allow healthcare professionals to visualize internal structures and organs. This visualization helps in the accurate diagnosis of various medical conditions, including fractures, tumors, infections, and cardiovascular diseases. For patient treatment planning, imaging provides essential information for planning and guiding medical interventions and surgeries. Surgeons can use preoperative imaging to understand the anatomy of the areas they will be operating on, resulting in more precise and effective procedures. Furthermore, medical imaging is crucial for monitoring disease progression and assessing treatment effectiveness. It allows healthcare providers to track changes in tumor size and characteristics, assess organ status, and adjust treatment plans accordingly. In some cases, medical imaging can detect disease at an early stage, before symptoms appear. Early diagnosis often leads to more successful treatment outcomes and can significantly improve patient prognosis.

[0025] Other affected areas include research and education. Medical imaging is crucial for medical research, allowing scientists and healthcare professionals to study diseases, develop new treatments, and enhance their understanding of human anatomy and physiology. It is also a valuable tool in medical education for training future healthcare professionals. In emergency situations, rapid and accurate diagnosis is critical, and imaging techniques such as X-rays and CT scans play a vital role in quickly assessing injuries and determining appropriate actions. Non-invasive assessment is another area where medical imaging can provide non-invasive techniques to visualize the body's internal structures, reducing the need for exploratory surgeries; this minimizes patient discomfort, shortens recovery time, and lowers healthcare costs. Overall, medical imaging is revolutionizing healthcare by providing valuable insights into the structure and function of the human body. Advances in imaging technologies continuously enhance diagnostic capabilities, improve treatment strategies, and ultimately contribute to better patient outcomes.

[0026] While the use of contrast agents in medical imaging can significantly enhance the visualization of organs and structures, potential image quality problems can also arise. It is important for healthcare professionals to understand these issues to ensure accurate interpretation of contrast-enhanced images. Some common image quality problems include artifacts, which are unwanted features or distortions in an image that can be caused by a variety of factors, including patient movement, metallic implants, or problems associated with contrast agent injection. Artifacts degrade image quality and affect the accuracy of interpretation. Inadequate contrast enhancement, where the contrast agent fails to provide sufficient enhancement of the target organ or lesion, can be due to factors such as insufficient contrast dosage, improper injection technique, or physiological changes in the patient. Timing issues associated with attempting to achieve optimal contrast enhancement typically require precise timing of the contrast agent injection relative to image acquisition. Variations in cycle time and injection rate can affect the quality of enhancement and result in suboptimal images. Some patients may experience allergic reactions to contrast agents. Although these reactions are relatively rare, they can range from mild to severe and can affect the imaging process.

[0027] Healthcare providers need to closely monitor patients for signs of allergic reactions. Contrast-induced nephropathy is a potential problem, especially with iodine-based contrast agents used in CT imaging. Patients with impaired kidney function may be at risk, and healthcare professionals carefully evaluate the risk-benefit ratio of contrast agent use in such cases. Another issue is that some lesions or tissues may exhibit heterogeneous enhancement patterns, making accurate image interpretation challenging. Radiologists need to consider the possibility of varying enhancement within the same organ or lesion. Variability in patient characteristics, such as body size, can affect contrast agent distribution, thus impacting image quality. Special considerations may be required for pediatric or obese patients. Problems with imaging equipment or technical parameters can affect image quality. Regular quality assurance checks and calibrations are crucial to ensuring optimal performance. Post-processing of contrast-enhanced images can introduce artifacts. Radiologists need to be aware of potential artifacts and carefully examine images to distinguish between genuine pathology and artifacts. Integrating contrast-enhanced images with non-contrast images or images from other modalities can sometimes be challenging. Co-registration issues can affect the accurate relevance of findings. Continuous efforts are needed to address and mitigate these image quality problems. Radiologists and imaging technicians undergo training to optimize imaging protocols, minimize artifacts, and accurately interpret images. Furthermore, technological advancements and the development of new contrast agents aim to improve the overall quality and safety of contrast-enhanced imaging.

[0028] The novelty of this innovation lies in its being an AI-based CT image enhancement technique for different organs / ROIs that automatically assigns optimal brightness and contrast to provide consistent and accurate interpretation of organs, contours, edges, and transition areas. This innovation features automated organ-specific brightness and contrast enhancement derived from training using initial segmentation. It provides automatic enhancement for different ROIs, thus offering a corresponding organ-specific optimized view of the input image.

[0029] The advantages of neural networks are utilized for enhancement via a fully AI-based approach; consequently, organ / ROI edges are smoother and more consistent due to the characteristics of neural networks and extensive training. This innovation builds upon a previously proposed invention, the so-called remapping algorithm, which predicts window level (WL) and window width (WW) maps instead of a direct image-to-image translation. Therefore, automatically adjusted WW / WL can be obtained and analyzed from the mapping, making it easier for clinicians to interpret AI results.

[0030] Figure 1 A block diagram of an exemplary non-limiting system 100 for improving deep learning image processing according to one or more embodiments described herein is shown. As shown, the image processing system 100 can be electronically integrated with medical images 110 via any suitable wired or wireless electronic connection.

[0031] In various embodiments, medical image 110 can depict any suitable anatomical structure of any suitable medical patient. As some non-limiting examples, the anatomical structure can be any suitable tissue of the medical patient (e.g., bone tissue, lung tissue, muscle tissue, brain tissue), any suitable organ of the medical patient (e.g., heart, liver, lungs, brain, eyes, colon, blood vessels), any suitable body fluid of the medical patient (e.g., blood, amniotic fluid), any other suitable body part of the medical patient, or any suitable portion thereof.

[0032] In all aspects, the medical image 110 may exhibit any suitable format, size, or dimension. As a non-limiting example, for any suitable positive integers x and y, the medical image 110 may be an x×y pixel array. As another non-limiting example, for any suitable positive integers x, y, and z, the medical image 104 may be an x×y×z voxel array.

[0033] In various instances, the medical image 110 may be generated or otherwise captured by any suitable medical imaging device, medical imaging apparatus, or medical imaging modality (not shown). As a non-limiting example, the medical image 110 may be generated or otherwise captured by a CT scanner, in which case the medical image 110 may be considered a CT scan image. As another non-limiting example, the medical image 110 may be generated or otherwise captured by an MRI scanner, in which case the medical image 110 may be considered an MRI scan image. As yet another non-limiting example, the medical image 110 may be generated or otherwise captured by a PET scanner, in which case the medical image 110 may be considered a PET scan image. As yet another non-limiting example, the medical image 110 may be generated or otherwise captured by an X-ray scanner, in which case the medical image 110 may be considered an X-ray scan image. As yet another non-limiting example, the medical image 110 may be generated or otherwise captured by an ultrasound scanner, in which case the medical image 110 may be considered an ultrasound scan image. In addition, medical images 110 can undergo appropriate image reconstruction techniques, such as filtered back projection.

[0034] In various embodiments, the medical image 110 can be considered as consisting of multiple regions. In other words, the pixels or voxels of the medical image 110 can be considered as being assigned, distributed, or otherwise divided among these multiple regions such that each region can be considered a subset of the pixels or voxels of the medical image. In various aspects, each pixel or voxel can be assigned to one (only one) of these multiple regions. Furthermore, in various instances, each region can have two or more pixels or voxels assigned to it. Therefore, the cardinality of these multiple regions can be less than the cardinality of pixels or voxels in the medical image 110. In other words, fewer regions than pixels or voxels can exist.

[0035] In various embodiments, the image processing system 100 may include a processor 102 (e.g., a computer processing unit, microprocessor) and a non-transitory computer-readable storage device 104 operatively or communicatively connected to or coupled to the processor 102. The non-transitory computer-readable storage device 104 may store computer-executable instructions that, when executed by the processor 102, cause the processor 102 or other components of the image processing system 100 (e.g., receiving component 106, AI model optimization component 108) to perform one or more actions. In various embodiments, the non-transitory computer-readable storage device 104 may store computer-executable components (e.g., receiving component 106, AI model optimization), and the processor 102 may execute these computer-executable components.

[0036] In various embodiments, the image processing system 100 may include a receiving component 106. In various aspects, the receiving component 106 may electronically receive or otherwise electronically access images of a set of "regions of interest / volumes" containing multiple organs. In various instances, the receiving component 106 may electronically retrieve the medical image 110 from any suitable centralized or distributed data structure (not shown) or from any suitable centralized or distributed computing device (not shown). As a non-limiting example, any medical imaging device, apparatus, or modality that generates or captures the medical image 110 (e.g., a CT scanner, MRI scanner, X-ray scanner, PET scanner, ultrasound scanner) may transmit the medical image 110 to the receiving component 106. In any case, the receiving component 106 may electronically acquire or access the medical image 110, enabling other components of the image processing system 100 to electronically interact with the medical image 110.

[0037] In various embodiments, the image processing system 102 may include an AI model optimization component 108. In various aspects, as described herein, the artificial intelligence deep learning neural network model component automatically processes and enhances the corresponding image in a locally adaptive manner, such that at each location, the enhanced image is optimized for the organ displayed at that location, and the deep learning neural network can be performed on the medical image 110.

[0038] The various embodiments described herein can be considered as computerized tools (e.g., any suitable combination of computer-executable hardware or computer-executable software) that can facilitate improved deep learning image processing. In various aspects, such computerized tools may include access components, inference components, transformation components, or display components.

[0039] In various embodiments, the medical image may depict one or more anatomical structures (e.g., tissues, organs, body parts, or portions thereof) of a medical patient (e.g., a person, animal, or other). In various instances, the medical image may exhibit any suitable size, format, or dimension (e.g., it may be a two-dimensional pixel array or a three-dimensional voxel array). In various cases, the medical image may be generated or otherwise captured by any suitable medical imaging modality (e.g., via a CT scanner, via an MRI scanner, via an X-ray scanner, via a PET scanner, or via an ultrasound scanner). In various aspects, the medical image may have undergone any suitable image reconstruction technique (e.g., filtered backprojection).

[0040] In any case, the pixels or voxels of the medical image can be considered as being allocated, divided, or otherwise distributed among multiple regions. In various aspects, each given pixel or voxel of the medical image can be assigned to only one of these multiple regions. Furthermore, in various instances, each given region can have two or more pixels or voxels assigned to it. Therefore, the total number of regions can be less than the total number of pixels or voxels in the medical image. In other words, each of the multiple regions can be a strict subset of the pixels or voxels of the medical image, each strict subset having a cardinality greater than 1, and wherein such strict subsets do not intersect each other (e.g., do not overlap).

[0041] In various aspects, a region of the medical image can be considered as any suitable contiguous cluster of pixels or voxels. That is, more than one pixel or voxel can be assigned to the region, and any two pixels or voxels assigned to the region can be: adjacent to each other (e.g., in a row and a column of each other); or coupled to each other through an unbroken chain of other pixels or voxels, wherein each pixel or voxel in such an unbroken chain is also assigned to the region, and wherein each contiguous pair of pixels or voxels in such an unbroken chain is adjacent to each other. In various other aspects, a region of the medical image can be non-contiguous. That is, a region can be considered as two or more distinct contiguous clusters of pixels or voxels that are spatially separated or otherwise do not contact each other. In other words, there can be two pixels or voxels assigned to the region that are: not adjacent to each other; and not coupled to each other through any unbroken chain of other contiguous adjacent pixels or voxels also assigned to the region. In various cases, some regions of the plurality of regions can be contiguous, while others of the plurality of regions can be non-contiguous.

[0042] In various aspects, the regions of the medical image can exhibit any suitable shape (if the region is continuous) or multiple shapes (if the region is non-continuous). For example, a region may comprise more than one pixel or voxel of the medical image, which may collectively form (due to their spatial location or orientation within the medical image) any suitable regular or convex polygonal shape (e.g., a square, a rectangle). In other instances, a region may comprise more than one pixel or voxel of the medical image, which may collectively form (due to their spatial location or orientation within the medical image) any suitable irregular or non-convex polygonal shape. In various cases, different regions may have the same or different shapes from each other.

[0043] In various aspects, regions can be defined by any suitable interval or range of pixel intensity values or voxel intensity values. For example, for any suitable positive real numbers a < b < c, all pixels of the medical image having intensity values greater than or equal to a and less than b can belong to one region of the plurality of regions, and all pixels of the medical image having intensity values greater than or equal to b and less than c can belong to another region of the plurality of regions. In various cases, different regions can be defined by distinct, different, non - overlapping, or otherwise non - overlapping intensity intervals or ranges.

[0044] In various aspects, regions can be defined based on the tissue types depicted in the medical image or otherwise based on the tissue types depicted in the medical image. For example, assume the medical image shows d distinct tissue types (e.g., bone tissue, lung tissue, skin tissue, skeletal muscle tissue, cardiac muscle tissue) for any suitable positive integer d > 1. In this case, the plurality of regions can include d + 1 regions: one distinct region for each of the d distinct tissue types, and one distinct region for all pixels or voxels that do not belong to all of the d distinct tissue types. In various instances, the d distinct tissue types can be identified in any suitable manner. For example, in some cases, a pre - trained tissue segmentation model can be executed on the medical image, and such a pre - trained tissue segmentation model can produce a segmentation mask as output that indicates which pixels or voxels of the medical image belong to which of the d distinct tissue types.

[0045] Regardless of their shape, regardless of their continuity, and regardless of how else they are defined (e.g., based on intensity intervals or tissue types), the plurality of regions can be considered to jointly form the medical image. That is, the union of the plurality of regions can be equal to the medical image. In other words, the plurality of regions can fit together like the tiles of a jigsaw puzzle to jointly produce the medical image.

[0046] In various embodiments, the receiving component 106 of system 100 can electronically receive or otherwise electronically access the medical image 110. In some aspects, the receiving component 106 can electronically retrieve the medical image 110 from any suitable centralized or decentralized data structure (e.g., a graphical data structure, a relational data structure, a hybrid data structure), whether remote from the receiving component 106 or local to the receiving component. For example, the receiving component 106 can retrieve the medical image 110 from any medical imaging device that generated or captured the medical image. In any case, the receiving component 106 can electronically obtain or access the medical image 110 such that other components of the computerized tool can electronically interact with the medical image 110 (e.g., read, write, edit, copy, manipulate).

[0047] In various implementations, the system's inference component 114 may electronically store, maintain, control, or otherwise access the deep learning neural network. In various cases, the deep learning neural network may exhibit any suitable internal architecture. For example, a deep learning neural network may include any suitable number of layers of any suitable type (e.g., an input layer, one or more hidden layers, an output layer, any of which may be a convolutional layer, a dense layer, a nonlinear layer, a pooling layer, a batch normalization layer, or a padding layer). As another example, a deep learning neural network may include any suitable number of neurons in various layers (e.g., different layers may have the same or different numbers of neurons). As yet another example, a deep learning neural network may include any suitable activation function (e.g., softmax, sigmoid, hyperbolic tangent, corrected linear unit) in various neurons (e.g., different neurons may have the same or different activation functions). As yet another example, a deep learning neural network may include any suitable inter-neuron or inter-layer connections (e.g., forward connections, skip connections, recursive connections).

[0048] In any case, as described herein, the deep learning neural network can be configured to receive a medical image as input and produce a region-by-region parameter mapping as output. Therefore, the inference unit can electronically execute the deep learning neural network on the medical image to produce a set of region-by-region parameter mappings corresponding to the medical image. More specifically, the inference unit 114 can feed the medical image 110 into the input layer of the deep learning neural network, which can perform forward propagation of one or more hidden layers of the deep learning neural network, and the output layer of the deep learning neural network can compute the set of region-by-region parameter mappings based on the activations generated by the one or more hidden layers.

[0049] In various aspects, this set of region-by-region parameter mappings may include any suitable number of region-by-region parameter mappings. In various instances, a region-by-region parameter mapping may be any suitable electronic data (e.g., vector, matrix, tensor) that indicates, specifies, or otherwise represents parameters (e.g., scalar coefficients with any suitable magnitude) corresponding respectively (e.g., in a one-to-one manner) to the plurality of regions. In other words, a region-by-region parameter mapping may contain, include, or encompass a unique or distinct parameter (e.g., a scalar coefficient) for each unique or distinct region of the medical image, or may consist of the aforementioned parameters. In contrast to a pixel-by-pixel or voxel-by-voxel parameter mapping, which would instead indicate, specify, or otherwise represent parameters corresponding respectively to pixels or voxels of the medical image (e.g., the pixel-by-pixel or voxel-by-voxel parameter mapping would instead have a unique or distinct parameter for each unique or distinct pixel or voxel of the medical image), this region-by-region parameter mapping may be used to indicate, specify, or otherwise represent parameters corresponding respectively to pixels or voxels of the medical image. That is, compared with the pixel-by-pixel or voxel-by-pixel parameter mapping corresponding to the medical image, the region-by-region parameter mapping corresponding to the medical image can have fewer parameters (e.g., it can have a smaller dimension).

[0050] As a non-limiting example, suppose the medical image is a 200×200 pixel array, and suppose each disjoint 10×10 pixel block is considered a region of the medical image. In this case, the medical image could have a total of 40,000 pixels (e.g., 200×200=40,000), and the medical image could have a total of 400 regions (e.g., since each region can be a disjoint 10×10 pixel block, there could be 20 rows and 20 columns of such 10×10 blocks in the medical image; 20×20=400). Accordingly, the pixel-wise parameter mapping in this case would be a vector, matrix, or tensor with 40,000 parameters (e.g., one parameter per pixel), while the region-wise parameter mapping in this case would be a vector, matrix, or tensor with only 400 parameters (e.g., one parameter per region). That is, in this non-limiting example, such a region-wise parameter mapping could have two orders of magnitude fewer parameters than the pixel-wise parameter mapping.

[0051] Because a region-by-region parameter mapping corresponding to the medical image can have fewer parameters than a pixel-by-pixel or voxel-by-pixel parameter mapping corresponding to the medical image (e.g., in some cases, by several orders of magnitude), this set of region-by-region parameter mappings can also collectively have fewer parameters than a similar set of pixel-by-pixel or voxel-by-pixel parameter mappings (e.g., in some cases, by several orders of magnitude). Accordingly, since the deep learning neural network can be configured to output this set of region-by-region parameter mappings, it can have fewer layers or neurons compared to the case where the deep learning neural network is alternatively configured to generate a similar set of pixel-by-pixel or voxel-by-pixel parameter mappings (e.g., in some cases, by several orders of magnitude). In other words, by configuring the deep learning neural network to output this set of region-by-region parameter mappings instead of a similar set of pixel-by-pixel or voxel-by-pixel parameter mappings, the footprint of the deep learning neural network can be reduced (e.g., in some cases, by several orders of magnitude).

[0052] In various implementations, the transformation component 116 of the computerized tool 100 can electronically generate a transformed version of the medical image by feeding the medical image and the set of region-by-region parameter maps into an analytical transformation function. More specifically, the analytical transformation function may include any suitable number of any suitable type of mathematical operators (e.g., polynomial operators, logarithmic operators, exponential operators, trigonometric operators) that can be combined in any suitable manner (e.g., multiplication or addition). Regardless of the specific mathematical operators implemented, the analytical transformation function can take any given pixel or voxel of the medical image as a variable parameter, and a parameter from each of the set of region-by-region parameter maps corresponding to any region to which the given pixel or voxel is assigned as a variable parameter, and the analytical transformation function can produce a transformed pixel or transformed voxel as output. In other words, the analytical transformation function can update the value of a pixel or voxel based on the current value of any pixel or voxel of the medical image and also based on a parameter from each of the set of region-by-region parameter maps. In any case, the analytical transformation function can be applied to each pixel or voxel of the medical image in this manner to produce a transformed version of the medical image. Note that this allows the transformed version of the medical image to have the same format, size, or dimension as the original medical image (e.g., the same number or arrangement of pixels or voxels). In other words, the analytical transformation function can change the intensity values ​​of pixels or voxels in the medical image, but can keep the position, orientation, or arrangement of the pixels or voxels unchanged.

[0053] In various embodiments, the display component of the computerized system 100 can electronically render a transformed version of the medical image on any suitable electronic display (e.g., a computer screen, computer monitor, graphical user interface). Therefore, a user, technician, or medical professional can visually examine or observe the transformed version of the medical image rendered on the electronic display, which can assist in diagnosis or prognosis. Furthermore, in various aspects, the display component 120 can electronically render any one of the group of region-by-region parameter maps on the electronic display. Therefore, the user, technician, or medical professional can visually examine such region-by-region parameter maps, which can also assist in diagnosis or prognosis.

[0054] To help ensure accurate or correct mapping of the region-by-region parameters, the deep learning neural network can first undergo training of any suitable type or paradigm (e.g., supervised training, unsupervised training, reinforcement learning). Therefore, in various aspects, receiving component 106 can receive, retrieve, or access the training dataset, and computerized system 100 may include a training component (not shown) capable of training the deep learning neural network on that training dataset.

[0055] In various aspects, the training dataset may include multiple training medical images. In various instances, the training medical images may have the same size, format, or dimension (e.g., the same number or arrangement of pixels or voxels) as the medical images discussed above. For example, if the medical image is a two-dimensional pixel array depicting the anatomical structure of a medical patient, then each training medical image may also be a two-dimensional pixel array depicting the corresponding anatomical structure of that medical patient. As another example, if the medical image is a three-dimensional voxel array depicting the anatomical structure of a medical patient, then each training medical image may also be a three-dimensional voxel array depicting the corresponding anatomical structure of that medical patient.

[0056] In any case, each training medical image can be considered to correspond to multiple regions discussed above. That is, the pixels or voxels of each training medical image can be considered to be allocated, divided, or otherwise assigned to the same number of regions as the pixels or voxels of the medical images discussed above. For example, suppose the pixels or voxels of the medical images discussed above are allocated to h regions for any suitable positive integer h > 1. In this case, the pixels of each voxel of the training medical image can be allocated to these h regions in the same way. That is, the medical image can be considered to have a first region, and each training medical image can also be considered to have a corresponding first region. Similarly, the medical image can be considered to have an h-th region, and each training medical image can also be considered to have a corresponding h-th region.

[0057] Note that, in some instances, a region of any given training medical image can have pixels or voxels that are located in the same positions as the corresponding regions of the medical images discussed above. For example, if for any suitable positive integers g ≤ h, i < j, and k < l, the g-th region of a medical image includes pixels located in the i-th to j-th rows and k-th to l-th columns of the medical image, then the g-th region of a given training medical image can similarly include pixels located in the i-th to j-th rows and k-th to l-th columns of that given training medical image.

[0058] However, note that, in some instances, a region of any given training medical image can have pixels or voxels that are not located in the same positions as the corresponding regions of the medical images discussed above. For example, if the g-th region of the medical image includes all pixels of the medical image within a given intensity interval, then the g-th region of the given training medical image can similarly include all pixels of that given training medical image within the given intensity interval. However, because the intensity value distribution of the medical image may be different from the intensity distribution of the given training medical image, the pixels of the given training medical image that fall within the given intensity interval may be in different positions compared to the pixels of the medical image that fall within the given intensity interval. In fact, a different number of pixels of the given training medical image may fall within the given intensity interval compared to the pixels of the medical image.

[0059] As another example, if the g-th region of the medical image includes all pixels of the medical image that belong to a particular tissue type, then the g-th region of the given training medical image can similarly include all pixels of that given training medical image that belong to the particular tissue type. However, because the anatomical structure depicted in the medical image may be different from the anatomical structure depicted in the given training medical image, the pixels of the given training medical image that belong to the particular tissue type may be in different positions compared to the pixels of the medical image that belong to the particular tissue type. In fact, a different number of pixels of the given training medical image may belong to the particular tissue type compared to the pixels of the medical image.

[0060] In any case, the training dataset can include the plurality of training medical images. In various aspects, the training dataset can include multiple sets of true region-by-region parameter maps that respectively correspond to the plurality of training medical images. In various instances, a set of true region-by-region parameter maps can have the same size, format, or dimension as the set of region-by-region parameter maps discussed above. In other words, a set of true region-by-region parameter maps can be considered to indicate or represent the correct or accurate region-by-region parameter maps that are known or believed to correspond to the respective training medical images.

[0061] In various cases, the training dataset may include multiple real-world transformed medical images, each corresponding to one of the multiple training medical images. In each aspect, the real-world transformed medical images may have the same size, format, or dimension (e.g., the same number or arrangement of pixels or voxels) as the transformed versions of the medical images discussed above. In other words, each real-world transformed medical image may be considered as an indication or representation of a transformed version (e.g., a tissue-equalized version, a brightness / contrast-enhanced version, a denoised version, a modal-modified version) of a corresponding training medical image that is known or considered correct or accurate.

[0062] In each aspect, the training component can perform supervised training on the deep learning neural network based on the training dataset. Before such supervised training begins, the trainable intrinsic parameters of the deep learning neural network (e.g., weights, biases, convolutional kernels) can be randomly initialized.

[0063] In various aspects, the training component can select any suitable training medical image from the training dataset, the true region-by-region parameters corresponding to any suitable set of such selected training medical images, and any suitable true transformed medical image corresponding to such selected training medical images. In various cases, the training component can feed the selected training medical image into a deep learning neural network, which can cause the deep learning neural network to produce a first output. For example, the training component can feed the training medical image into the input layer of the deep learning neural network, which performs forward propagation of one or more hidden layers of the deep learning neural network, and the output layer of the deep learning neural network can compute the first output based on the activations from the one or more hidden layers. Note that in various cases, the size, format, or dimension of the first output can be controlled or otherwise determined by the number or arrangement of neurons in the output layer (e.g., the first output can be forced to have a desired size, format, or dimension by adding neurons to or removing neurons from the output layer of the deep learning neural network).

[0064] In all respects, the first output can be viewed as the set of predicted or inferred region-by-region parameter mappings that the deep learning neural network believes should correspond to the selected training medical images. Conversely, the selected set of true region-by-region parameter mappings can be considered as correct or accurate region-by-region parameter mappings that are known or otherwise considered to correspond to the selected training medical images. Note that if the deep learning neural network has not undergone training or has undergone very little training so far, the first output may be extremely inaccurate (e.g., it may differ significantly from the selected set of true region-by-region parameter mappings).

[0065] In various instances, the training component can generate a second output by feeding both the selected training medical image and the first output into the analysis transformation function. More specifically, the second output can be considered a predicted or inferred transformed version of the selected training medical image (e.g., a predicted or inferred tissue equalization version, a predicted or inferred brightness-contrast enhancement version, a predicted or inferred denoising version, a predicted or inferred modality modification version). Conversely, the selected real transformed medical image can be considered a correct or accurate transformed version (e.g., a correct or accurate tissue equalization version, a correct or accurate brightness-contrast enhancement version, a correct or accurate denoising version, a correct or accurate modality modification version) that is known or otherwise considered to correspond to the selected training medical image. Note that if the deep learning neural network has not undergone training or has undergone very little training so far, the second output may be extremely inaccurate (e.g., it may differ significantly from the selected real transformed medical image).

[0066] In each respect, the training component can compute any suitable error or loss (e.g., mean absolute error (MAE), mean squared error (MSE), cross entropy) between the first output and the selected set of ground truth region-by-region parameter mappings. Similarly, the training component can compute any suitable error or loss (e.g., MAE, MSE, cross entropy) between the second output and the selected ground truth transformed medical image. Accordingly, the training component can update the trainable intrinsic parameters (e.g., weights, biases, convolutional kernels) of the deep learning neural network by performing backpropagation (e.g., stochastic gradient descent) driven by such computed errors or losses.

[0067] In various instances, this supervised training process can be repeated for each training medical image in the training dataset, resulting in the trainable intrinsic parameters of the deep learning neural network being iteratively optimized to accurately generate region-by-region parameter mappings for the input medical images. In various cases, this training component can implement any suitable training batch size, any suitable training termination criterion, or any suitable error, loss, or objective function.

[0068] Note that in some aspects, the training dataset may lack the multiple sets of true region-by-region parameter mappings. In this case, the selected set of true region-by-region parameter mappings may not exist, but the selected true transformed medical images may still exist. Therefore, backpropagation can be driven by any calculated error or loss between the second output and the selected true transformed medical images. Conversely, note that in other aspects, the training dataset may lack the multiple true transformed medical images. In this case, the selected true medical images may not exist, but the selected set of true region-by-region parameter mappings may still exist. Therefore, backpropagation can be driven by any calculated error or loss between the first output and the selected set of true region-by-region parameter mappings (e.g., in this case, the analytical transformation function may not need to be applied during training).

[0069] The various embodiments described herein can be used to solve inherently highly technical problems (e.g., to facilitate improved deep learning image processing) using hardware or software, problems that are not abstract and cannot be performed as a set of human mental behaviors. Furthermore, some processes performed can be executed by a dedicated computer (e.g., a deep learning neural network with trainable internal parameters such as convolutional kernels) to implement defined actions related to improved deep learning image processing. For example, such defined actions may include: a device operatively coupled to a processor accessing a medical image, wherein the pixels or voxels of the medical image are allocated across multiple regions; generating a set of region-wise parameter maps by the device and via the execution of the deep learning neural network on the medical image, wherein the region-wise parameter maps consist of a predicted parameter for each region of the medical image; the device generating a transformed version of the medical image by feeding the set of region-wise parameter maps into an analytical transformation function; and rendering the transformed version of the medical image on an electronic display.

[0070] Such defined actions are not performed manually by humans. In fact, neither human thought nor a person holding pen and paper can electronically access medical images (e.g., CT scans, MRI scans, X-ray scans), electronically execute deep learning neural networks on those images to generate region-by-region parameter maps (as opposed to pixel-by-pixel or voxel-by-voxel parameter maps), electronically use those region-by-region parameter maps to generate a transformed version of the medical image, and electronically display that transformed version of the medical image on a computer screen. Similarly, deep learning neural networks are inherently computerized constructs that cannot be realized in any way by human thought without a computer. Likewise, medical images are essentially computerized constructs generated or captured by electronic medical hardware (e.g., CT scanners, MRI scanners, X-ray scanners, PET scanners, ultrasound scanners), not generated in any way by human thought without the use of computers. Accordingly, the computerized tools that can train or execute deep learning neural networks to produce transformed versions of medical images are also inherently computerized and cannot be implemented in any sensible, practical, or reasonable way without a computer.

[0071] Furthermore, the various embodiments described herein can control real-world tangible devices based on the disclosed teachings. For example, the various embodiments described herein can electronically execute (or train) real-world deep learning neural networks on real-world medical images (e.g., CT images, MRI images, X-ray images, PET images, ultrasound images), and can electronically render any results produced by such real-world deep learning neural networks on a real-world computer screen.

[0072] It should be understood that the accompanying figures and descriptions provide non-limiting examples of various implementations and are not necessarily drawn to scale. Furthermore, the terms AI / NN model deep learning optimization component, AI deep learning, AI neural network, neural network, and deep learning are used interchangeably but reflect the same concepts.

[0073] Figure 2 Examples of various images with different brightness and contrast according to one or more embodiments described herein are illustrated. The individual images (202, 204, 206, 208) reveal how broader enhancement can reveal more detail and define structures with greater clarity. However, since one image may focus on a single organ, it may reduce the quality of the rest of the content when analyzing the image.

[0074] Figure 3An example of an image optimization process according to one or more embodiments described herein is illustrated. For this innovation, the basic process begins with image 302, which is sent to model 304 to initiate transformation. Any segmentation is avoided at inference time and is only considered for annotation during the training phase. Therefore, annotation work is minimized to define only preferred organ appearances, e.g., WL (window level) / WW (window width) per organ. Furthermore, since the AI ​​model generates per-pixel maps in a regression manner rather than using segmentation masks, training the AI ​​model is easier than with conventional segmentation-based methods. The predicted WW / WL 306 is the output and is compared with the ground truth WL GT 308 and WW GT 310 to measure the difference and loss function. The remapping function 318 is based on a nonlinear transformation and provides a prediction of the per-organ WW / WL based on parameters that can regenerate the original image into the desired image. The comparison between the prediction 316 and the enhanced GT 312 image is the result of the loss function 314 determining acceptability. For this innovation, the inventors map inputs to functions and functions to outputs. Having intermediate functions allows the process to reduce the artifact problem that direct mapping would introduce.

[0075] Optimizing parameters is also important in optimizing organ imaging. In organ imaging systems, the kilovolt (kV) setting plays a crucial role in determining image quality and patient dose. High kV settings typically range from 120 kV to 140 kV and offer advantages such as reduced patient dose because higher kV settings require less radiation to penetrate the body. Another benefit is better penetration, which is very effective for imaging larger patients. Low kV settings typically range from 80 kV to 100 kV and provide better differentiation between tissues, especially useful for detecting small or subtle lesions. The choice between high and low kV settings depends on the specific diagnostic task, patient size, and the organ being imaged. For example, low kV settings are often used for chest X-rays to enhance contrast, while high kV settings may be preferred for abdominal CT scans to reduce dose. In organ imaging, milliampere-seconds (mA) are a key parameter affecting image quality and patient dose. The mA value is the product of tube current (in milliamperes (mA)) and exposure time (in seconds). A higher mA increases the number of X-ray photons, resulting in better image quality with less noise. However, it also increases the radiation dose to the patient. While a lower mA reduces the number of X-ray photons, which can produce a noisier image, it reduces the radiation dose to the patient. The goal is to find the optimal balance between image quality and patient dose. This typically involves adjusting the mA setting based on the specific imaging task, patient size, and the organ being imaged. In organ imaging, particularly in CT (computed tomography) scans, pitch is also a crucial parameter affecting both image quality and patient dose. Pitch is defined as the ratio of the stage travel distance during a complete gantry rotation to the total width of the X-ray beam collimator 1. The optimal pitch depends on the specific imaging task and the balance between image quality and patient dose. Gantry rotation speed is a critical parameter in CT imaging, affecting both image quality and patient dose. The speed at which the CT scanner gantry rotates around the patient affects image quality. Faster gantry rotation speeds can improve temporal resolution and reduce motion artifacts, especially when imaging moving organs like the heart.

[0076] Figure 4Example non-limiting block diagrams illustrating how an AI / NN deep learning neural network can be trained on a training dataset with reference to a benchmark ground truth are shown according to one or more embodiments described herein. An important aspect of this innovation is the quality of the training data and the training process itself. A lack of rigorous training will result in poor image outcomes and may lead to significant delays in accuracy. This can be corrected with new, comprehensive data for training, but this is resource-intensive and time-consuming. In all aspects, before starting such training, the training algorithm 408 is able to initialize the trainable intrinsic parameters (e.g., convolutional kernels, weight matrices, bias values) of the AI / NN deep learning neural network 108 in any suitable manner (e.g., random initialization).

[0077] In various instances, the training component 408 can select from the most recent training dataset of the training medical image 402, a set of true region-by-region parameter mappings 404 corresponding to the training medical image 402, and the true transformed medical image 406 corresponding to the training medical image 402. A "ground truth" refers to definitive and accurate information about a patient's condition or the characteristics of a particular medical image. It serves as a reference standard against which the performance of an imaging algorithm or diagnostic method is evaluated. Ground truths are typically established through rigorous and reliable means, such as histopathological examination, surgical findings, or long-term clinical follow-up. In various cases, as shown, the training component 408 can be executed on the AI / NN deep learning neural network 108 on the training medical image 402, causing the AI / NN deep learning neural network 108 to produce an output 410. More specifically, the training component 408 can feed the training medical image 402 into the input layer of the AI / NN deep learning neural network 108, the training medical image 402 can perform forward propagation to one or more hidden layers of the AI / NN deep learning neural network 108, and the output layer of the AI / NN deep learning neural network 108 can compute the output 410 based on the activations generated by one or more hidden layers of the AI / NN deep learning neural network 108.

[0078] Note that, in various cases, the size, format, or dimension of output 410 can be controlled or otherwise determined by the number of neurons arranged in the output layer of the AI / NN deep learning neural network 108 (or by the characteristics of other internal parameters such as convolutional kernels). That is, by controllably adding neurons (or other internal parameters such as convolutional kernels) to the output layer of the AI / NN deep learning neural network 108, removing neurons (or other internal parameters such as convolutional kernels) from the output layer of the AI / NN deep learning neural network, or adjusting neurons (or other internal parameters such as convolutional kernels) in the output layer of the AI / NN deep learning neural network, output 410 can be forced to have a desired size, format, or dimension.

[0079] In each respect, output 410 can be viewed as p predicted or inferred region-by-region parameter mappings that the AI / NN deep learning neural network 108 believes should correspond to the training medical image 402. Conversely, the set of true region-by-region parameter mappings 404 can be viewed as p known or believed to correspond to the training medical image 402 correctly or accurately. Note that if the AI / NN deep learning neural network 108 has not undergone training or has undergone very little training so far, this output 410 may be extremely inaccurate (e.g., it may differ greatly from the set of true region-by-region parameter mappings 404).

[0080] In various instances, training algorithm 408 can feed output 410 and training medical image 402 to analysis transformation algorithm 412 to produce output 414. Accordingly, output 414 can be considered a predicted or inferred transformed version of training medical image 402 (e.g., a predicted or inferred tissue equalization version, a predicted or inferred brightness-contrast enhancement version, a predicted or inferred denoising version, a predicted or inferred modality modification version). Conversely, the real transformed medical image 406 can be considered a known correct or accurate transformed version of training medical image 402 (e.g., a correct or accurate tissue equalization version, a correct or accurate brightness-contrast enhancement version, a correct or accurate denoising version, a correct or accurate modality modification version). As mentioned above, note that if AI / NN deep learning neural network 108 has not undergone or has undergone very little training so far, output 414 may be highly inaccurate (e.g., may be very different from the real transformed medical image 406).

[0081] In various aspects, training component 408 can compute any suitable error or loss (e.g., MAE, MSE, cross-entropy) between output 410 and the set of true region-by-region parameter mappings 404. Similarly, in various instances, training component 408 can compute any suitable error or loss (e.g., MAE, MSE, cross-entropy) between output 414 and the true transformed medical image 406. In various cases, training component 408 can incrementally update the trainable intrinsic parameters (e.g., convolutional kernels, weights, biases) of AI / NN deep learning neural network 108 via backpropagation based on such computed errors or losses.

[0082] In various aspects, the training component 408 can repeat this execution and update process for each training medical image in the training dataset. This can ultimately lead to the iterative optimization of the trainable intrinsic parameters (e.g., convolutional kernels, weights, biases) of the AI / NN deep learning neural network 108 to accurately generate region-by-region parameter mappings for the input medical images. In various instances, the training component 408 can implement any suitable training batch size, any suitable training termination criterion, or any suitable error, loss, or objective function.

[0083] In some cases, the training dataset may lack the multiple sets of true region-by-region parameter mappings 404. In such cases, it can be essentially as follows: Figure 4 The AI / NN deep learning neural network 108 is trained as described, except that the set of real region-by-region parameter maps 404 may not be available. In this case, the training algorithm 408 avoids calculating the error or loss between the output 410 and the set of real region-by-region parameter maps 404. However, the training component 408 can still calculate the error or loss between the output 414 and the real transformed medical image 406, and this error or loss can drive backpropagation.

[0084] In other cases, the training dataset may lack several real-world transformed medical images 406. In this case, an AI / NN deep learning neural network 202 can be trained, except that the real-world transformed medical images 406 may not be available. In this case, the training algorithm 408 can avoid calculating the error or loss between the output 414 and the real-world transformed medical images 406. However, the training component 408 can still calculate the error or loss between the output 410 and the set of real-world region-by-region parameter mappings 404, and backpropagation can be driven by this error or loss. Note that in this case, the training component 408 does not necessarily need to apply the analysis transformation function to the output 410. It should be noted that there are many ways to train an AI / NN deep learning system, and what is discussed are merely various examples and may or may not be part of an implementation scheme. The system is trained to fully understand what an organ is during the training process and does not play a role at inference time. The organ is understood by the system in terms of training time and optimized images. All segmentation occurs at training time, not at inference time.

[0085] Figure 5 Figure 7b illustrates examples of current segmentation-based images and images processed by the present invention according to one or more embodiments described herein. Figure 5 The enhanced image reveals incredible detail, particularly in the lower part of the image, highlighting the vertebral structure 504 and its relationship to adjacent anatomical structures—compared to the current segmentation-based image 502. Figure 6We observed a clear difference. The enhanced image 604 technique clearly reveals more detail than the original image 602, and is therefore more likely to reveal even subtle changes that might indicate early disease or abnormalities. Figures 7a and 7b contain two sets of images. Again, this comparison emphasizes a wider range of grayscale values. With more levels of enhancement in 704, and richer detail compared to the current solution 702, it can be used as a valuable tool for identifying early or subtle pathological findings. Figure 7b reflects similar results, as 708 presents a clearer image than 706.

[0086] Figure 8 A flowchart illustrating an image optimization process according to one or more embodiments described herein, and a non-limiting computer implementation of a method to facilitate enhanced image processing, is provided. The basic technical sequence begins with 802, where the first step is to create a standard image encoder, which can be any neural network model, such as the state-of-the-art SAM encoder. An encoder generally refers to a component or algorithm used to transform raw data or images into different representations. This process is typically part of the image reconstruction or processing pipeline in a medical imaging system. The term "encoder" can have different meanings depending on the specific imaging modality or technology. In some cases, encoders can be used for data compression, where raw image data is transformed into a more compact representation without losing critical information. This can be important for reducing storage requirements and accelerating the transfer of medical images.

[0087] An encoder can also be part of a feature extraction process as identified in 804. In this case, the encoder transforms the input data into a set of representative features that capture important information about the underlying structure in a medical image. This feature representation can be used for tasks such as image classification, segmentation, or other analyses. In medical imaging, encoders may be involved in signal processing tasks. For example, in computed tomography (CT) imaging, an encoder can be used to convert raw X-ray measurements into an image representation through processes such as backprojection and filtering. With the rise of deep learning and neural networks in medical imaging, the term "encoder" is often associated with the encoding portion of an autoencoder architecture. An autoencoder is a neural network consisting of an encoder and a decoder. The encoder compresses the input data into a latent spatial representation, and the decoder reconstructs the original input from this representation. It can be decoded by any decoder, such as UNet or a simple upsampling + convolution.

[0088] The decoder is the counterpart to the encoder. It takes the compressed or latent representation produced by the encoder and reconstructs the original input data. In the context of medical imaging, the decoder is designed to generate an output that is very similar to the input image.

[0089] At 806, the decoder predicts the WL / WW map for each pixel and transforms the input image based on the WL / WW map for each pixel. It is important to note that this functionality was implemented in a previous "remapping" invention, and this innovation is built upon that. The specific role and function of the encoder can vary depending on the imaging modality (e.g., CT, MRI, ultrasound) and the specific task or application within medical imaging. At 808, the loss function between the desired image and the predicted image is calculated. The desired WL / WW GT is obtained based on the ROI segmentation mask and a predefined WL / WW for each organ. The loss function (also called the cost function or objective function) is a measure of the difference between the predicted output and the actual target (in this case, the image). The goal during neural network training is to minimize this loss function. This is a crucial area and is primarily based on training. Another possible option to consider is photon counting computed tomography (PCCT), a form of X-ray computed tomography (CT) where X-rays are detected using a photon counting detector (PCD) that records the interactions of individual photons. By tracking the deposited energy in each interaction, the detector pixels of the PCD record an approximate energy spectrum, making it a spectral or energy-resolved CT technique.

[0090] At 810, a comparison is made between the predicted image and the actual image, and based on the result, additional training may be required at 812 or the image may be acceptable at 816.

[0091] Figure 9 Example 900 illustrates an image optimization process using text as a cue to guide feature decoding according to one or more embodiments described herein. Training is as follows: a ground truth (GT) is generated based on the organ category of interest, such as text [lung, liver, kidney, or any potential organ]. The text is encoded by a pre-trained CLIP model for embedding feature extraction. The extracted linguistic features are then fused with image features for decoding. The generated image is learned using the GT, augmented by regions selected from the text, and the loss term can be a simple MAE loss.

[0092] While the content of this paper primarily describes various implementations applied to deep learning AI neural networks, these are merely non-limiting examples. In all respects, the teachings described herein can be applied to any suitable machine learning model exhibiting any appropriate artificial intelligence architecture (e.g., Support Vector Machine, Naive Bayes, Linear Regression, Logistic Regression, Decision Tree, Random Forest).

[0093] In various instances, machine learning algorithms or models may be implemented in any suitable manner to facilitate any suitable aspect described herein. To facilitate some of the machine learning aspects described above in various implementations, consider the following discussion of artificial intelligence (AI). The various implementations described herein may employ artificial intelligence to facilitate the automation of one or more features or functionalities. These components may employ various AI-based schemes to perform the various implementations / examples disclosed herein. To provide or contribute to the numerous determinations described herein (e.g., determination, detection, inference, accounting, prediction, prognosis, estimation, derivation, forecasting, detection, computation), the components described herein may examine the entirety or a subset of the data to which they have been granted access and may provide inference about or determine the state of a system or environment from a set of observations captured via events or data. For example, determination may be used to identify a specific context or action, or to generate a probability distribution of states. These determinations may be probabilistic; that is, the calculation of the probability distribution of states of interest is based on consideration of data and events. Determination may also refer to techniques used to compose higher-level events from a collection of events or data.

[0094] Such determinations can lead to the construction of new events or actions from observed events or a collection of stored event data, regardless of whether the events are closely related in time, and regardless of whether the events and data come from one or more event and data sources. The components disclosed herein can be combined to perform automatic or deterministic actions related to the claimed subject matter using various classification schemes or systems (e.g., support vector machines, neural networks, expert systems, Bayesian confidence networks, fuzzy logic, data fusion engines, etc.) that are explicitly trained (e.g., via training data) and implicitly trained (e.g., via observed behavior, preferences, historical information, received external information, etc.)). Therefore, classification schemes or systems can be used for automatic learning and the performance of multiple functions, actions, or determinations.

[0095] The classifier can take the input attribute vector z=(z1, z2, z3, z4, z) as an input vector. nThis maps the input to a confidence level that it belongs to a certain category, as shown by f(z) = confidence level (category). Such classification can employ probability-based or statistical analysis (e.g., analyzing utility and cost considerations) to determine the action to be automated. Support Vector Machines (SVMs) are an example of a usable classifier. SVMs operate by finding a hypersurface in the space of possible inputs, where the hypersurface attempts to separate triggering criteria from non-triggering events. Intuitively, this makes the classification correct for test data that is close to but different from the training data. Other directed and undirected model classification methods include, for example, Naive Bayes, Bayesian networks, decision trees, neural networks, fuzzy logic models, or any of the probabilistic classification models that provide different independent patterns. Classification, as used in this paper, also includes statistical regression for developing priority models.

[0096] The disclosure herein describes non-limiting examples. For ease of description or explanation, various parts of the disclosure herein use the terms "each," "every," or "all" when discussing various examples. Such use of the terms "each," "every," or "all" is non-limiting. In other words, when the disclosure herein provides a description applicable to "each," "every," or "all" of a particular object or component, it should be understood that this is a non-limiting example, and it should also be understood that in various examples, such a description may apply to fewer than "each," "every," or "all" of that particular object or component.

[0097] To provide additional context for the various implementation schemes described herein, Figure 10 The following discussion is intended to provide a brief general description of a suitable computing environment 1000 in which various implementations of the embodiments described herein may be implemented. While the implementations have been described above in the general context of computer-executable instructions that can run on one or more computers, those skilled in the art will recognize that these implementations may also be implemented in combination with other program modules or as a combination of hardware and software.

[0098] Typically, program modules include routines, programs, components, data structures, etc., that perform specific tasks or implement specific abstract data types. Furthermore, those skilled in the art will understand that the methods of this invention can be practiced with other computer system configurations, including single-processor or multi-processor computer systems, minicomputers, mainframe computers, Internet of Things (IoT) devices, distributed computing systems, and personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, each operatively coupled to one or more associated devices.

[0099] The embodiments illustrated in this paper can also be practiced in a distributed computing environment where a specific task is performed by a remote processing device linked via a communication network. In a distributed computing environment, program modules can reside on both local and remote memory storage devices.

[0100] Computing devices typically include a variety of media, which may include computer-readable storage media, machine-readable storage media, or communication media, wherein the two terms are used interchangeably herein, as described below. A computer-readable storage medium or a machine-readable storage medium can be any available storage medium accessible by a computer, and includes volatile and non-volatile media, removable and non-removable media. By way of example and not limitation, a computer-readable storage medium or a machine-readable storage medium can be implemented in conjunction with any method or technology used for storing information, such as computer-readable or machine-readable instructions, program modules, structured data, or unstructured data.

[0101] Computer-readable storage media may include, but is not limited to, random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, optical disc read-only memory (CDROM), digital versatile disc (DVD), Blu-ray disc (BD) or other optical disc storage devices, magnetic tape cassettes, magnetic tape, disk storage devices or other magnetic storage devices, solid-state drives or other solid-state storage devices, or other tangible or non-transitory media that can be used to store desired information. In this regard, the terms “tangible” or “non-transitory” used herein to describe storage devices, memories, or computer-readable media shall be understood to exclude only the propagation of transient signals themselves as a modifier, and shall not waive any rights enjoyed in respect of all standard storage devices, memories, or computer-readable media other than the propagation of transient signals themselves.

[0102] Computer-readable storage media can be accessed by one or more local or remote computing devices, for example, via access requests, queries or other data retrieval protocols, to perform various operations with respect to the information stored on the media.

[0103] Communication media typically contain computer-readable instructions, data structures, program modules, or other structured or unstructured data in a data signal, such as a modulated data signal, a carrier wave, or other transmission mechanism, and include any information transmission or delivery medium. The terms "modulated data signal" or "signal" refer to a signal whose one or more characteristics are set or altered to encode information in one or more signals. By way of example and not limitation, communication media include wired media (such as wired networks or direct wired connections) and wireless media (such as acoustic, RF, infrared, and other wireless media).

[0104] Refer againFigure 10 Example environment 1000 for implementing various embodiments of the aspects described herein includes a computer 1002, which includes a processing unit 1004, system memory 1006, and a system bus 1008. The system bus 1008 couples system components, including but not limited to the system memory 1006, to the processing unit 1004. The processing unit 1004 can be any of a variety of commercially available processors. Dual microprocessors and other multiprocessor architectures may also be used as the processing unit 1004.

[0105] System bus 1008 can be any of several types of bus structures that can be further interconnected to a memory bus (with or without a memory controller), a peripheral bus, and a local bus using any of a variety of commercially available bus architectures. System memory 1006 includes ROM 1010 and RAM 1012. The basic input / output system (BIOS) can be stored in non-volatile memory such as ROM, erasable programmable read-only memory (EPROM), or EEPROM, where the BIOS contains basic routines that facilitate, for example, transferring information between components within computer 1002 during startup. RAM 1012 may also include high-speed RAM, such as static RAM for caching data.

[0106] Computer 1002 also includes an internal hard disk drive (HDD) 1014 (e.g., EIDE, SATA), one or more external storage devices 1016 (e.g., floppy disk drive (FDD) 1016, memory stick or flash drive reader, memory card reader, etc.), and drives 1020 (e.g., solid-state drives, optical disc drives) capable of reading from or writing to disks 1022 (such as CD-ROMs, DVDs, BDs, etc.). Alternatively, in cases involving solid-state drives, disks 1022 are not included unless separate. Although the internal HDD 1014 is illustrated as being located within computer 1002, the internal HDD 1014 may also be configured for use outside of suitable infrastructure (not shown). Additionally, although not shown in environment 1000, solid-state drives (SSDs) may be used as a supplement to or replacement for HDD 1014. HDD 1014, external storage device 1016, and drive 1020 can be connected to system bus 1008 via HDD interface 1024, external storage interface 1026, and drive interface 1028, respectively. Interface 1024 for the specific implementation of the external drive may include at least one or both of Universal Serial Bus (USB) and Institute of Electrical and Electronics Engineers (IEEE) 13104 interface technologies. Other external drive connection technologies are contemplated in the embodiments described herein.

[0107] The drive and its associated computer-readable storage medium provide non-volatile storage of data, data structures, computer-executable instructions, etc. For computer 1002, the drive and storage medium are adapted to store any data in a suitable digital format. Although the above description of computer-readable storage media refers to a corresponding type of storage device, those skilled in the art will understand that other types of computer-readable storage media (whether currently existing or developed in the future) can also be used in the example operating environment, and furthermore, any such storage medium may contain computer-executable instructions for performing the methods described herein.

[0108] Multiple program modules may be stored in the drive and RAM 1012, including an operating system 1030, one or more application programs 1032, other program modules 1034, and program data 1036. All or part of the operating system, application programs, modules, or data may also be cached in RAM 1012. The systems and methods described herein can be implemented using various commercially available operating systems or combinations of operating systems.

[0109] Computer 1002 may optionally include emulation technology. For example, a hypervisor (not shown) or other intermediate device may emulate the hardware environment of operating system 1030, and the emulated hardware may optionally be different from the hardware of operating system 1030. Figure 10 The hardware illustrated herein. In this implementation, operating system 1030 may include one of a plurality of virtual machines (VMs) hosted at computer 1002. Furthermore, operating system 1030 may provide a runtime environment, such as the Java Runtime Environment or the .NET Framework, to application 1032. A runtime environment is a consistent execution environment that allows application 1032 to run on any operating system that includes that runtime environment. Similarly, operating system 1030 may support containers, and application 1032 may be in the form of a container of lightweight, standalone, executable software packages, including, for example, application code, runtime, system tools, system libraries, and settings.

[0110] Furthermore, computer 1002 can utilize security modules such as Trusted Processing Modules (TPMs) for enabling this. For example, in the case of a TPM, the boot part is hashed in the next boot part and waits for the result to match a security value before loading the next boot part. This process can occur at any layer of the computer 1002's code execution stack, such as at the application execution level or the operating system (OS) kernel level, thereby achieving security at any code execution level.

[0111] Users can input commands and information into computer 1002 through one or more wired / wireless input devices (e.g., keyboard 1038, touchscreen 1040, and pointing devices such as mouse 1042). Other input devices (not shown) may include microphones, infrared (IR) remote controls, radio frequency (RF) remote controls or other remote controls, joysticks, virtual reality controllers or virtual reality headsets, gamepads, styluses, image input devices (e.g., cameras), gesture sensor input devices, visual motion sensor input devices, emotion or face detection devices, biometric input devices (e.g., fingerprint or iris scanners), etc. These and other input devices are often connected to processing unit 1004 through input device interface 1044, which may be coupled to system bus 1008, but these and other input devices may also be connected through other interfaces (e.g., parallel ports, IEEE 13104 serial ports, game ports, USB ports, IR interfaces, BLUETOOTH). ® (Interfaces, etc.) connections.

[0112] Monitor 1046 or other types of display devices can also be connected to system bus 1008 via an interface such as video adapter 1048. In addition to monitor 1046, the computer typically includes other peripheral output devices (not shown), such as speakers, printers, etc.

[0113] Computer 1002 can operate in a networked environment using a logical connection to one or more remote computers (such as remote computer 1050) via wired or wireless communication. Remote computer 1050 can be a workstation, server computer, router, personal computer, portable computer, microprocessor-based entertainment device, peer-to-peer device, or other public network node, and typically includes many or all of the elements described relative to computer 1002, but for simplicity, only memory / storage device 1052 is illustrated. The depicted logical connection includes wired / wireless connectivity to a local area network (LAN) 1054 or a larger network (e.g., a wide area network (WAN) 1056). Such LAN and WAN networking environments are common in offices and companies and facilitate enterprise-wide computer networks (such as intranets), all of which can connect to global communication networks (e.g., the Internet).

[0114] When used in a LAN networking environment, computer 1002 can connect to local network 1054 via a wired or wireless communication network interface or adapter 1058. Adapter 1058 facilitates wired or wireless communication with LAN 1054, which may also include a wireless access point (AP) configured thereon for communication with adapter 1058 in wireless mode.

[0115] When used in a WAN networking environment, computer 1002 may include modem 1060, or may be connected to a communication server on WAN 1056 via other components (such as via the Internet) for establishing communication over WAN 1056. Modem 1060, which may function as an internal or external device and as a wired or wireless device, may be connected to system bus 1008 via input device interface 1044. In a networking environment, program modules depicted relative to computer 1002 or parts thereof may be stored in remote memory / storage device 1052. It should be understood that the network connections shown are examples, and other components for establishing communication links between computers may be used.

[0116] When used in a LAN or WAN networking environment, computer 1002 can access cloud storage systems or other network-based storage systems, such as, but not limited to, network virtual machines that provide one or more aspects of information storage or processing, as a supplement to or alternative to external storage device 1016 as described above. Typically, the connection between computer 1002 and the cloud storage system can be established, for example, via adapter 1058 or modem 1060 through LAN 1054 or WAN 1056. When computer 1002 is connected to the associated cloud storage system, external storage interface 1026 can manage the storage provided by the cloud storage system by means of adapter 1058 or modem 1060, just like other types of external storage devices. For example, external storage interface 1026 can be configured to provide access to cloud storage sources as if those cloud storage sources were physically connected to computer 1002.

[0117] Computer 1002 is operable to communicate with any wireless device or entity located wirelessly, such as a printer, scanner, desktop or portable computer, portable data assistant, communications satellite, any equipment or location associated with a wirelessly detectable tag (e.g., telephone booth, newsstand, store shelf, etc.), and telephone. This may include Wi-Fi and Bluetooth. ® Wireless technology. Therefore, communication can be a predefined structure like a regular network, or simply self-organizing communication between at least two devices.

[0118] Figure 11This is a schematic block diagram of a sample computing environment 1100 to which the disclosed subject matter can interact. The sample computing environment 1100 includes one or more clients 1110. Clients 1110 can be hardware or software (e.g., threads, processes, computing devices). The sample computing environment 1100 also includes one or more servers 1130. Servers 1130 can also be hardware or software (e.g., threads, processes, computing devices). For example, server 1130 can accommodate threads to perform transformations by employing one or more embodiments as described herein. One possible communication between client 1110 and server 1130 can be in the form of data packets suitable for transmission between two or more computer processes. The sample computing environment 1100 includes a communication framework 1150 that can be used to facilitate communication between client 1110 and server 1130. Client 1110 is operatively connected to one or more client data repositories 1120, which can be used to store information local to client 1110. Similarly, server 1130 is operatively connected to one or more server data repositories 1140, which can be used to store information locally on server 1130.

[0119] This invention can be a system, method, apparatus, or computer program product at any possible level of integrated technical detail. A computer program product may include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to execute aspects of the invention. The computer-readable storage medium may be a tangible device that holds and stores instructions for use by an instruction execution device. The computer-readable storage medium may be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. A less complete list of more specific examples of computer-readable storage media may also include: portable computer floppy disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital versatile disk (DVD), memory sticks, floppy disks, mechanical encoding devices (such as punch cards or protrusions in grooves on which instructions are recorded), and any suitable combination of the foregoing. As used herein, a computer-readable storage medium should not be construed as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses passing through fiber optic cables), or electrical signals transmitted through wires.

[0120] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to a corresponding computing / processing device, or downloaded to an external computer or external storage device via a network (e.g., the Internet, a local area network, a wide area network, or a wireless network). The network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them for storage in a computer-readable storage medium within the corresponding computing / processing device. The computer-readable program instructions used to perform the operations of this invention may be assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages ​​(including object-oriented programming languages ​​such as Smalltalk, C++, etc.) and procedural programming languages ​​such as the "C" programming language or similar programming languages. Computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet through an Internet service provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs) may execute computer-readable program instructions to personalize the electronic circuitry for performing aspects of the invention by utilizing state information of the computer-readable program instructions.

[0121] This document describes aspects of the invention with reference to flowchart illustrations or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations or block diagrams, and combinations of blocks in the flowchart illustrations or block diagrams, can be implemented by computer-readable program instructions. These computer-readable program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create components for implementing the functions / actions specified in one or more blocks of the flowchart or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that instructs a computer, programmable data processing apparatus, or other device to function in a particular manner, such that the computer-readable storage medium having the instructions stored therein includes an article of manufacture comprising instructions that implement aspects of the functions / actions specified in one or more blocks of the flowchart or block diagram. Computer-readable program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operations to be performed on the computer, other programmable apparatus, or other device to produce a computer-implemented process, such that the instructions, which execute on the computer, other programmable apparatus, or other device, implement the functions / actions specified in one or more blocks of the flowchart or block diagram.

[0122] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible specific implementations of systems, methods, and computer program products according to various embodiments of the invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of instructions comprising one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions indicated in the blocks may not occur in the order shown in the figures. For example, two blocks shown consecutively may actually be executed substantially simultaneously, or sometimes they may be executed in reverse order, depending on the functionality involved. It will also be noted that each block illustrated in a block diagram or flowchart, and combinations of blocks illustrated in a block diagram or flowchart, may be implemented by a dedicated hardware-based system that performs the specified function or action or implements a combination of dedicated hardware and computer instructions.

[0123] Although the subject matter has been described above in the general context of computer-executable instructions of a computer program product running on one or more computers, those skilled in the art will recognize that this disclosure may also be implemented in combination with other program modules. Typically, program modules include routines, programs, components, data structures, etc., that perform a specific task or implement a specific abstract data type. Furthermore, those skilled in the art will understand that other computer system configurations can be used to practice the computer implementation of the present invention, including single-processor or multi-processor computer systems, small computing devices, mainframe computers, and computers, handheld computing devices (e.g., PDAs, telephones), microprocessor-based or programmable consumer or industrial electronic devices, etc. The illustrated aspects can also be practiced in a distributed computing environment in which tasks are performed by remote processing devices linked via a communication network. However, some (if not all) aspects of this disclosure can be practiced on a standalone computer. In a distributed computing environment, program modules may reside on both local and remote memory storage devices.

[0124] As used herein, the terms “component,” “system,” “platform,” “interface,” etc., may refer to or include computer-related entities or entities associated with an operator having one or more specific functionalities. Entities disclosed herein may be hardware, a combination of hardware and software, software, or software in execution. For example, a component may be, but is not limited to, a process, processor, object, executable file, execution thread, program, or computer running on a processor. By way of example, both an application running on a server and the server itself can be components. One or more components may reside within a process or execution thread, and components may be located on a single computer or distributed among two or more computers. In another example, a corresponding component may execute on various computer-readable media on which various data structures are stored. Components may communicate via local or remote processes, such as based on signals having one or more data packets (e.g., data from one component that interacts with another component in a local system, a distributed system, or a network (such as the Internet with other systems) via signals). As another example, a component may be a device having specific functionalities provided by mechanical parts operated by electrical or electronic circuitry, which is operated by software or firmware applications executed by a processor. In such cases, the processor may be internal or external to the device and may execute at least a portion of a software or firmware application. As another example, a component may be a means of providing specific functionality through electronic parts rather than mechanical components, wherein the electronic parts may include a processor or other components for executing software or firmware that at least partially endows the electronic parts with functionality. In one aspect, a component may be emulated, for example, via a virtual machine within a cloud computing system.

[0125] Furthermore, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or". That is, unless otherwise specified or explicitly stated from the context, "X adopts A or B" is intended to mean any natural inclusive permutation. That is, if X adopts A; X adopts B; or X adopts both A and B, then "X adopts A or B" is satisfied in any of the foregoing cases. As used herein, the term "and / or" is intended to have the same meaning as "or". Furthermore, unless otherwise specified or explicitly stated from the context as to be directed to the singular form, the articles "a" and "an" used in this specification and figures should generally be understood to mean "one or more". As used herein, the terms "example" or "exemplary" are used to mean used as an example, instance, or illustration. For the avoidance of doubt, the subject matter disclosed herein is not limited to such examples. Additionally, any aspect or design described herein as an "example" or "exemplary" is not necessarily to be construed as preferred or advantageous over other aspects or designs, nor does it exclude equivalent exemplary structures and techniques known to those skilled in the art.

[0126] As used herein, the term "processor" can refer substantially to any computing processing unit or device, including but not limited to a single-core processor; a single processor with software multithreading capabilities; a multi-core processor; a multi-core processor with software multithreading capabilities; a multi-core processor with hardware multithreading technology; a parallel platform; and a parallel platform with distributed shared memory. Additionally, a processor can refer to an integrated circuit, an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a field-programmable gate array (FPGA), a programmable logic controller (PLC), a complex programmable logic device (CPLD), discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. Furthermore, processors can utilize nanoscale architectures (such as, but not limited to, molecular and quantum dot-based transistors, switches, and gates) to optimize space usage or enhance the performance of user equipment. Processors can also be implemented as a combination of computing processing units. In this disclosure, terms such as "repository," "storage device," "data repository," "data storage device," "database," and substantially any other information storage component related to the operation and functionality of a component are used to refer to a "memory component," an entity embodied in "memory," or a component that includes memory. It should be understood that the memory or memory component described herein may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. By way of illustration and not limitation, non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), flash memory, or non-volatile random access memory (RAM) (e.g., ferroelectric RAM (FeRAM)). For example, volatile memory may include RAM that can act as external cache memory. By way of illustration and not limitation, RAM can be provided in a variety of forms, such as synchronous RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), direct Rambus RAM (DRRAM), direct Rambus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM). Furthermore, the memory components disclosed in the systems or computer-implemented methods herein are intended to include, but are not limited to, these and any other suitable types of memory.

[0127] The foregoing description includes only examples of systems and computer-implemented methods. Of course, it is impossible to describe every conceivable combination of components or computer-implemented methods for the purposes of describing this disclosure, but many other combinations and arrangements are possible. Furthermore, when the terms “comprising,” “having,” or “possessing” are used in the detailed description, claims, appendices, and drawings, such terms are intended to have a similar inclusiveness to “comprising” as a transitional term in the claims.

[0128] Various embodiments have been described for illustrative purposes, but these descriptions are not intended to be exhaustive or limited to the disclosed embodiments. Many modifications and variations will be apparent without departing from the scope and substance of the described embodiments. The terminology used herein is chosen to best illustrate the principles of the embodiments, the practical application of or improvement of technology available on the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A system comprising: A processor that executes computer-executable components stored in a non-transitory computer-readable memory, wherein the computer-executable components include: A receiving component receives a set of "regions of interest / volumes" images containing multiple organs; and An artificial intelligence deep learning neural network model component automatically processes and enhances corresponding images in a locally adaptive manner, such that at each location, the enhanced image is optimized for the organ displayed at that location.

2. The system of claim 1, wherein the artificial intelligence deep learning neural network model component predicts window level (WL) and window width (WW) maps for each of the plurality of identified scanned organs.

3. The system according to claim 1, wherein the artificial intelligence deep learning neural network model uses regression techniques to predict WL and WW graphs.

4. The system of claim 1, wherein the artificial intelligence deep learning neural network model is trained in part by calculating a loss function between the desired organ image and the predicted organ image.

5. The system of claim 1, wherein the images of the plurality of scanned organs of the plurality of organs are obtained using at least one of the following: single-energy computed tomography (CT) or dual-energy computed tomography (CT) or photon-counting computed tomography (PCCT).

6. The system of claim 4, wherein the artificial intelligence deep learning neural network model generates organ-specific optimized views of the corresponding image in part based on contrast levels and processing parameters.

7. The system of claim 1, wherein the artificial intelligence deep learning neural network model generates organ-specific views of the corresponding image in part based on a comparison with a benchmark ground truth image.

8. The system of claim 7, wherein the reference truth image is generated using optimized acquisition parameters (high KV setting and low KV setting, mAmps, pitch, rack rotation speed or WL and WW views).

9. The system of claim 6, wherein the artificial intelligence deep learning neural network model employs one or more remapping algorithms to generate organ-specific optimized views of the corresponding images to smooth out and preserve details of the corresponding organ structures at region transitions.

10. A computer-implemented method, the method comprising: The processor, which executes computer-executable components stored in non-transitory computer-readable memory, performs the following actions: Receive a set of "region of interest / volume" images containing multiple organs; and An artificial intelligence deep learning neural network model is employed, which automatically processes and enhances the corresponding image in a locally adaptive manner, such that at each location, the enhanced image is optimized for the organ displayed at that location.

11. The method of claim 10, further comprising using an artificial intelligence deep learning neural network model component to predict window level (WL) and window width (WW) maps for each of a plurality of identified scanned organs.

12. The method of claim 10, further comprising using the artificial intelligence deep learning neural network model to predict WL and WW graphs using regression techniques.

13. The method of claim 10, further comprising training the artificial intelligence deep learning neural network model in part by calculating a loss function between the desired organ image and the predicted organ image.

14. The method of claim 10, further comprising obtaining images of the plurality of organs by using at least one of the following: single-energy computed tomography (CT), dual-energy computed tomography (CT), or photon-counting computed tomography (CT).

15. The method of claim 10, further comprising using the artificial intelligence deep learning neural network model to generate organ-specific optimized views of the corresponding image, in part based on contrast levels and processing parameters.

16. The method of claim 10, further comprising using the artificial intelligence deep learning neural network model to generate organ-specific views of the corresponding images in part based on comparisons with a benchmark ground truth image.

17. The method of claim 16, further comprising generating the reference truth image using optimized acquisition parameters (high KV setting and low KV setting, mAmps, pitch, rack rotation speed, or WL and WW views).

18. The method of claim 16, wherein the artificial intelligence deep learning neural network model employs one or more remapping algorithms to generate organ-specific optimized views of the corresponding images to smooth transitions between regions and preserve details of the corresponding organ structures.

19. A non-transitory machine-readable storage medium including executable instructions, said executable instructions facilitating the execution of operations when executed by a processor, said operations including: The processor receives a set of "regions of interest / volumes" images containing multiple organs; as well as The processor employs an artificial intelligence deep learning neural network model component that automatically processes and enhances the corresponding image in a locally adaptive manner, such that at each location, the enhanced image is optimized for the organ displayed at that location.

20. The non-transitory machine-readable storage medium of claim 19, the operation further comprising using an artificial intelligence deep learning neural network model to predict window level (WL) and window width (WW) maps for each of a plurality of identified scanned organs.

Images