Smart protocols for efficiently reviewing tomosynthesis data

Abstract

Various methods and systems are provided to reduce the time and resources spent viewing digital breast tomosynthesis (DBT) images and corresponding synthetic 2D images. [Solution] In one example, the method includes receiving multiple tomosynthesis data studies, each study including a first set of composite two-dimensional (S2D) images, a second set of slab images, and a third set of cross-sectional images. For each of the multiple tomosynthesis data studies, the minimum dataset to be viewed by the radiologist is determined based on the analysis of the tomosynthesis data by an AI-based system, and the minimum dataset, rather than the multiple tomosynthesis data studies, is displayed on the display device.

Description

Technical Field

[0001] Embodiments of the subject matter disclosed herein relate to imaging procedures for digital mammography.

Background Art

[0002] Mammography can obtain two-dimensional (2D) images of a subject's breast. Digital breast tomosynthesis (DBT) can obtain a three-dimensional (3D) image volume of a subject's breast, which is achieved by three-dimensionally reconstructing the breast from a plurality of 2D projections along different directions of the breast region. 3D modeling is typically performed by thick cutouts or thin cutouts (corresponding to relatively thick slices of the breast being examined and slices of zero thickness of the breast being examined, respectively). After the image volume is reconstructed, a radiologist reviews the 2D images corresponding to the image volume according to a predetermined reading protocol. However, in the reading protocol, it may be specified to review and analyze a large number of images, which takes time for the radiologist and may consume computing resources and memory resources that should otherwise be available for other tasks. In many cases, a diagnosis can be made based on a small number of images.

Summary of the Invention

[0003] In one embodiment, a protocol generation system for a digital mammography system includes a processor and a memory for storing instructions, the memory which, when the instructions are executed, causes the processor to receive a patient's digital breast tomosynthesis (DBT) case image, wherein the DBT case image includes at least one set of projection images and / or image volumes of the patient's breast reconstructed from data acquired using a digital mammography system; generate a case score for the DBT case file by analyzing the case image; select the minimum dataset from the DBT case images to be read based on the case score; and send the minimum dataset and / or a reading protocol specifying the minimum dataset, along with a dataset display sequence, to a radiology workstation (RWS) for reading, and / or save the minimum dataset to an archive.

[0004] It should be understood that the above summary is provided in a simplified form to introduce selected concepts from those further described in the embodiments for carrying out the invention. The above summary is not intended to identify any important or essential features of the claimed subject matter, and the scope of the claimed subject matter is uniquely defined by the claims. Furthermore, the claimed subject matter is not limited to implementations that solve the defects pointed out in the above portion or any portion of this disclosure. [Brief explanation of the drawing]

[0005] This disclosure will be further understood by reading the following description of non-limiting embodiments with reference to the accompanying drawings. [Figure 1] This is a schematic diagram of a digital mammography system according to one embodiment. [Figure 2] This figure shows a workflow according to embodiments disclosed herein, which allows for the selective transmission of a portion of a DBT study to a radiologist's workstation for review. [Figure 3] This flowchart shows a method for selectively transmitting a portion of a DBT study to a radiologist's workstation according to embodiments disclosed herein. [Figure 4] This flowchart shows a method for selecting portions of a DBT study received from a digital mammography system to be reviewed by a radiologist, according to embodiments disclosed herein. [Figure 5] This flowchart shows a method for encoding the minimum dataset to be reviewed from a DBT study and describing it in a Digital Image and Communications in Medicine (DICOM) tag for the DBT study, according to embodiments disclosed herein. [Figure 6] This flowchart shows a method for generating a minimum dataset for review from a DBT study using full-field digital mammography (FFDM) images, according to embodiments disclosed herein. [Figure 7] This flowchart shows a method for selectively reading a portion of a DBT study from an archive, according to embodiments disclosed herein. [Figure 8] This is an exemplary imaging system infrastructure used to carry out the methods shown in Figures 3 to 7 according to embodiments disclosed herein. [Figure 9] As prior art, an exemplary representation of a full DBT image dataset is shown. [Figure 10] This specification illustrates an exemplary representation of a breast density image displayed based on an image interpretation protocol selectively generated by a protocol generation system, according to embodiments disclosed herein. [Modes for carrying out the invention]

[0006] The following description relates to various embodiments of digital mammography imaging procedures. Mammography is a medical imaging procedure for detecting one or more tumors in the breast. Based on mammography imaging, a breast biopsy procedure may be performed to obtain a biopsy sample of the breast tissue in question for further analysis. During the breast biopsy procedure, the breast is compressed with compression paddles and positioned in either a medial-lateral or craniocaudal location, depending on whether the biopsy is performed using a horizontal approach (an approach in which the needle is inserted into the tissue along a medial-lateral plane parallel to the detector) or a vertical approach (an approach in which the needle is inserted vertically along a craniocaudal plane). The location of the target tissue (e.g., lesion, microcalcification, etc.) is then identified based on a mammography imaging procedure such as digital breast tomosynthesis (DBT). For example, during DBT, a scout image (with the X-ray tube positioned perpendicular to the detector) and multiple stereo images (when the X-ray tube moves in an arc at various angles in both positive and negative directions within a set angular range from the centerline) are obtained. The target location within the region of interest (ROI) can be selected based on the acquired images. Once the target is selected, a needle is inserted into the breast using a biopsy tool, and a portion of the target tissue is extracted to collect a biopsy sample.

[0007] Digital mammography imaging procedures can include acquiring two-dimensional (2D) or three-dimensional digital images of the breast. For example, DBT is an imaging technique that produces cross-sectional images of the breast with high in-plane resolution. During acquisition using a digital mammography system, the breast is compressed, and the X-ray source can rotate around the breast within positive and negative angular ranges from an intermediate position. Low-dose X-ray projection images of the breast at each angle can be obtained by the detector. The projection images are then reconstructed as slice images of the breast volume along the z-direction.

[0008] Some medical procedures (such as breast biopsies) may be performed with the assistance of contrast-enhanced imaging performed on a digital mammography system. Contrast-enhanced imaging involves administering a contrast agent (such as iodine) to the subject being imaged (e.g., the patient). The contrast agent flows through the patient's blood vessels, which can help visualize the biopsy target (e.g., lesion). After contrast agent administration, dual-energy images can be obtained at various points in the biopsy procedure (e.g., immediately after contrast agent injection and before anesthesia administration, after anesthesia administration, after biopsy needle insertion, after biopsy device operation, after sample collection, and / or after biopsy clip insertion). Dual-energy images can be generated from two images, including a first image acquired at low radiation energy (called a low-energy image or LE) and a second image acquired at high radiation energy (called a high-energy image or HE). Using digital subtraction, a dual-energy (DE) image can be generated from the LE and HE images, which removes background features from the DE image, allowing contrast-enhanced features (e.g., lesions) to be visualized more accurately.

[0009] During DBT, one or more two-dimensional (2D) images may be used to improve interpretation accuracy and for comparison with previous mammograms. 2D images may be generated using full-field digital mammography (FFDM), or using synthetic mammography (SM), which performs 2D reconstruction of tomosynthesis slice datasets to reduce image acquisition time and radiation exposure (compared to FFDM). The 3D image volume and 2D images generated from DBT can be interpreted by a radiologist on a flat screen based on an interpretation protocol. The interpretation protocol (also called a hanging protocol) can determine the display order of the images to be interpreted. For example, a synthetic 2D (S2D) mammogram of four standard screening views may be displayed first. Then, following user input, slabs for the four standard views are automatically displayed, followed by cross-sections. Such an interpretation protocol is sometimes called S2D+Cross-section+Slab.

[0010] However, S2D+section+slab interpretation protocols may require reviewing and analyzing a large number of images (for example, 244 images). Interpreting such a large number of images can be time-consuming for radiologists and may consume computational and memory resources that could have been used for other tasks. It is also possible to use different interpretation protocols, such as S2D+section protocols without slab interpretation, or S2D+slab protocols without section interpretation. However, even following these protocols may still result in reviewing a large amount of data. In some situations, if S2D (synthetic 2D) is sufficiently reliable to show diagnostically relevant information from the image volume in a single 2D image, an S2D-only interpretation protocol can be used without compromising the radiologist's performance. However, it is necessary to capture almost 100% of the diagnostic information from the image volume in the underlying S2D selection process, and setting the S2D selector's operating point to approximately 100% sensitivity may lead to a sharp increase in false positives and compromise S2D image quality.

[0011] To address these issues, this specification discloses a protocol generation system that employs a hybrid approach to protocol selection. The protocol generation system performs an evaluation of 3D image volumes and selectively provides one or more interpretation protocols based on the evaluation results. For cases that are easy to diagnose or for which automated analysis by AI means is highly reliable, the radiologist may be provided with an S2D-only protocol, while for difficult cases where automated analysis is less reliable, an S2D + slab / section protocol may be provided. Several embodiments are proposed, ranging from hybrid interpretation protocols to smart hanging protocols. In this way, the protocol generation system provides a way to automatically adjust the amount of data to be reviewed, thereby improving the overall workload of the radiologist. By adopting this hybrid approach, radiologists can reduce the time spent interpreting DBT images and synthetic 2D images without compromising interpretation performance. Reducing the time spent interpreting images reduces the amount of computational and memory resources used by the computing system for interpreting images.

[0012] Referring to the drawings, Figure 1 is a schematic diagram of a digital mammography system 100 for obtaining enhanced projection images of a site of interest (such as the breast), which includes an X-ray detector 145 and an X-ray source 140 facing it. The X-ray source 140 and the X-ray detector 145 are connected by an arm 144. A site of interest 132 can be placed between the X-ray detector 145 and the X-ray source 140. In the illustrated system, the X-ray source 140 moves in an arc above the X-ray detector 145. The X-ray detector 145 and multiple positions of the X-ray sources 140' and 140″ along the arc (see dashed line) are shown by the dashed / solid line and partial perspective view. In the illustrated configuration, the X-ray detector 145 is fixed in the illustrated position, and only the X-ray source 140 moves. However, in other embodiments, the X-ray detector 145 also moves, and the X-ray detector 145 follows the movement of the X-ray source 140, becoming substantially perpendicular to directions 141, 142, and 143. Angle α is the projection angle between the zero direction and other directions (such as 141 and 142). In this way, multiple different views of breast tissue can be obtained with at least one X-ray source 140. The projection of the smallest α, or the projection closest to the zero direction, is approximately called the central projection or zero projection.

[0013] Referring to Figure 1, the left side shows a partial perspective view of an imaging system according to an exemplary embodiment of the present disclosure, including an X-ray detector 145 and an X-ray source 140. Different positions of the X-ray source 140, 140', and 140'' are schematically shown to illustrate the movement of the X-ray source. Nine different projection views 101, 102, 103, 104, 106, 107, 108, and 109 are shown, including a zero projection (105) shown as a straight line, all of which point to the center of the X-ray detector 145.

[0014] The patient (not shown) is positioned in front of the digital mammography system. Taking medial-lateral oblique (MLO) acquisition or viewing as an example, the user or medical professional (e.g., mammography technician 181) sets the desired projection angle (30 to 60 degrees, where 45 degrees represents the preferred zero projection shown in the oblique view of Figure 1). For routine screening mammograms, an angled MLO view is preferred over a 90-degree lateral view because it allows for imaging of more breast tissue.

[0015] The region of interest 132 shown on the display unit 170 is the breast compressed by the compression paddles 133, which ensure uniform compression and fixation of the breast during radiation irradiation, resulting in optimal image quality. The breast 132 may include, for example, calcifications or lesions 131, which may be located in the zero direction 143 perpendicular to the plane of the detector 145. The user can review and diagnose calcifications, lesions, and other clinically relevant structures. The display unit 170 displays a 2D mammography image that primarily allows review of the central portion of the breast 132.

[0016] The X-ray detector 145 and the X-ray source 140 constitute an acquisition unit, which is connected to a processing unit 150 via a data acquisition line 155. The processing unit 150 has a memory unit 160, which can be connected to the processing unit 150 via an archive line 165.

[0017] A user or healthcare professional (such as a mammography technician 181) can input control signals through the user interface 180. These control signals are transmitted from the user interface to the processing unit 150 via the signal line 185. The method and system according to this disclosure allows a user to obtain enhanced 2D projection images similar to 2D mammograms. Based on these high-quality images, radiologists can identify all clinical signs relevant to breast examination. Users can easily analyze the displayed images, especially if they are familiar with 2D mammograms. Furthermore, previously acquired 2D mammograms may be displayed for comparison with images acquired with the tomosynthesis imaging system according to this disclosure. In addition, tomosynthesis images can be reviewed and archived. A CAD system or the user can affix 3D markers to the images. Height maps of clinically relevant structures or other objects obtained by embodiments of this disclosure can be combined with 3D marks from a CAD system or height information provided by the user through a 3D review system.

[0018] The memory unit 150 can be integrated into or separated from the processing unit 150. The memory unit 160 can store image data (such as 2D weighted projection images and tomosynthesis 3D images). Generally, the memory unit 160 may include a computer-readable medium (e.g., a hard disk or CD-ROM, floppy disk, ROM / RAM memory, DVD, a digital source such as a network or the internet, or other suitable means). The processing unit 150 is configured to execute program instructions stored in the memory unit 160, thereby causing the computer to perform the method of the present disclosure. One technical effect of carrying out the method according to embodiments of the present invention is that the number of X-ray source uses can be reduced. This is because weighted 2D projection images can be used instead of 2D mammograms, which typically require separate X-ray irradiation to obtain high-quality images.

[0019] After one or more DBT volumes and / or one or more mammograms are generated by the digital mammography system 100, those volumes and / or mammograms can be transmitted to a radiologist and interpreted. A set of 2D images can be obtained from these volumes and / or mammograms, and the radiologist interprets the set of 2D images according to an interpretation protocol. The interpretation protocol can specify the set of 2D images to be sent to the radiologist and can specify details on how to interpret the 2D images or in what order to interpret the 2D images. The interpretation protocol can be automatically selected by the digital mammography system 100 based on case criteria. The interpretation protocol can be an S2D + cross-section + slab interpretation protocol, which specifies that the radiologist interprets a synthetic 2D image generated from the DBT volume, a plurality of images corresponding to various slices of the DBT volume, and a plurality of images corresponding to various slabs (e.g., a plurality of adjacent slices) of a defined thickness of the DBT volume. In some examples, the interpretation protocol may be an S2D + cross-section interpretation protocol or an S2D + slab protocol. The radiologist can interpret the 2D images specified by the interpretation protocol and diagnose the patient.

[0020] Since it is considered that the total number of 2D images to be interpreted by the radiologist will be large and thus it is considered that interpreting the 2D images will take time, as will be described later with reference to FIGS. 2, 8, and 3, it is advantageous to use a protocol generation system to select an interpretation protocol, a hybrid interpretation protocol, or a plurality of interpretation protocols based on a specific patient case.

[0021] Referring now to FIG. 2, a simplified exemplary workflow 200 is shown for generating a reading protocol and a set of DBT / mammography images of a patient's breast for sending to a radiologist for reading. The DBT / mammography images can be generated by a digital mammography system 202 (such as the digital mammography system 100 of FIG. 1). The DBT volume reconstructed by the digital mammography system 202 can be input into an artificial intelligence (AI) model 204, which analyzes the volume to determine whether one or more tumors are present within the volume and further determines the characteristics of one or more tumors (e.g., extent, stage, whether the tumor is benign or malignant, type of tumor, etc.). The AI model can output a classification of the patient case. Based on the classification of the patient case, a reading protocol and a minimum data set of the images to be reviewed can be provided to the radiologist 212. For example, for a first classification indicating no findings, no minimum data set is generated and the 2D images of the patient case can be prevented from being sent to the radiologist's review workstation. In the case of a second classification indicating that a non-cancerous abnormality has been detected, a first reading protocol for reading a first minimum data set including S2D images 206 can be sent. In the case of a second classification indicating that there is a suspicion of a cancerous abnormality, a second reading protocol for reading a second minimum data set including S2D images 208 and a plurality of DBT images 210 is sent. In other embodiments, different reading protocols and different minimum data sets can be generated. The images sent to the radiologist 212 are reviewed by the radiologist 212 on a display screen 214 of the radiologist's computing device (such as a remote workstation (RWS)).

[0022] Figure 8 shows an infrastructure diagram 800 of exemplary infrastructure used to support workflow 200 and the workflows of other embodiments described herein. Infrastructure diagram 800 includes a digital mammography system 802, which is a non-limiting version of the digital mammography system 100 of Figure 1. The digital mammography system 802 can generate one or more images 804 of a patient's breast (including DBT volume images and / or 2D FFDM images). Images 804 may also include S2D images, and 2D images can be generated from DBT volumes by an S2D generator 803. The S2D generator 803 may be, or may include, an anomaly detection system used to generate a set of slabs, a set of planes, and / or S2D images from one or more images 804.

[0023] Image 804 can be transmitted to image analysis system 806. Image analysis system 806 may be integrated into digital mammography system 802 or may be provided on a separate computing device. Image analysis system 806 includes AI module 808, which includes multiple AI-based anomaly detection systems 810. Multiple AI-based anomaly detection systems 810 may include, for example, a risk assessment system; a negative triage system; a breast density assessment system; an anomaly detection system used to generate a set of slabs, a set of sections, and / or S2D images from DBT case images; and / or different anomaly detection systems.

[0024] One or more anomaly detection systems 810 can be used to process the image 804 and detect lesions, tumors, calcifications, and / or other anomalies. That is, one or more anomaly detection systems 810 can take image data of the image 804 as input and output encoded data of one or more findings, the findings may include detected lesions, tumors, etc. The findings may include, for example, the location of the anomaly in the image data; the classification of the anomaly; an assessment of the severity of the anomaly; and / or a confidence score indicating the confidence with which the anomaly detection system 810 assesses the severity of the anomaly.

[0025] Findings detected by one or more anomaly detection models 810 can be transmitted to a protocol generation system 820. In the embodiment shown in Figure 8, the protocol generation system 820 is installed in a separate computing device that is communicatively coupled to the image analysis system 806. In other embodiments, the protocol generation system 820 may be integrated into the digital mammography system 802 or into the image analysis system 806.

[0026] The protocol generation system 820 includes a processor 824 configured to execute machine-readable instructions stored in non-temporary memory 826. The processor 824 may be single-core or multi-core, and the program executed on the processor may be designed for parallel processing or for distributed processing. In some embodiments, the processor 824 may optionally include separate components distributed across two or more devices that can be located remotely and / or configured to perform cooperative processing. In some embodiments, one or more aspects of the processor 824 may be virtualized and implemented by a remotely accessible network computing device configured in a cloud computing configuration.

[0027] Non-temporary memory 826 includes a protocol generation model 830, which can be a classification model used to classify the output of the image analysis system 806. The protocol generation model 830 receives encoded data of one or more findings in image 804 and can output a classification of one or more findings. For example, if no findings are present, the protocol generation model 830 can output a first classification (e.g., 0). If a finding is detected and the finding can be diagnosed with high confidence (e.g., a benign tumor), the protocol generation model 830 can output a second classification (e.g., 1). If a finding is detected and the finding is serious and / or not easily diagnosed (e.g., a malignant tumor), the protocol generation model 830 can output a third classification (e.g., 2). In other embodiments, the protocol generation model 830 may output a larger or smaller number of classifications, or different types of classifications.

[0028] Based on the output of the protocol generation model 830, the protocol generation system 820 can generate an image interpretation protocol 840 for interpreting image 804. However, unlike typical image interpretation protocols for image 804, the image interpretation protocol 840 can specify a minimum dataset 842 (a dataset containing fewer images than image 804) that the radiologist should interpret. The minimum dataset 842 and the image interpretation protocol 840 are transmitted to the RWS 850 via a network (such as a wireless network 870) and can be displayed to the radiologist. Once the minimum dataset 842 and the image interpretation protocol 840 are received by the RWS 850, they are stored in the RWS 850's memory 854 and displayed on the RWS 850's display 856 based on the instructions contained in the image interpretation protocol 840 and executed by the RWS 850's processor 852. As detailed in Figures 3 to 8 below, the image interpretation protocol 840 has the advantage of reducing the number of images sent to the radiologist for interpretation, and can reduce the minimum dataset 842 in the network 870 and the bandwidth consumption of the image interpretation protocol 840.

[0029] In addition or alternatively, in some embodiments, the minimum dataset 842 and the image interpretation protocol 840 are transmitted to and stored in the archive 860 and can be retrieved from the archive 860 by the RWS 850. That is, the minimum dataset 842 and the image interpretation protocol 840 are not transmitted to the RWS 850, and can be retrieved from the archive 860 by the RWS 850. By retrieving the minimum dataset 842 and the image interpretation protocol 840 from the archive 860, a prefetch strategy can be employed to efficiently manage the bandwidth consumption of the minimum dataset 842 and the image interpretation protocol 840 on the network, as will be explained later with reference to Figure 8.

[0030] Referring to Figure 3, a flowchart of an exemplary method 300 for generating and transmitting an image interpretation protocol to a radiologist, following the workflow 200 described above, is shown. Method 300 can be executed using computer-readable instructions stored in the non-temporary memory of the computing device of a digital mammography system (e.g., digital mammography system 100 in Figure 1) or in a controller (e.g., controller 44 in Figure 1) that is communicatively coupled to the digital mammography system. In some embodiments, the protocol generation system can be installed and operated on another computing device (e.g., an edge device, an image archiving and communication system (PACS)) without departing from the scope of this disclosure.

[0031] Method 300 begins with step 302. Step 302 includes performing a DBT scan on a patient using a digital mammography system. An image volume can be reconstructed from projection data acquired during the DBT scan. For the purposes of this disclosure, patient data, image data, and the results of analyses performed on the image data can be collectively referred to as a patient case. Furthermore, S2D images can be generated from the image volume by various techniques known in the art. The S2D image generation process is based on automated analysis that searches for lesions in the DBT volume generated from the scan. Once the location of findings is detected, a rendering process generates a 2D image from the DBT acquired data that particularly highlights the findings.

[0032] In step 304, method 300 includes analyzing the image volume using one or more AI models (such as the anomaly detection model of the anomaly detection system 810 in Figure 8). Examples of AI models include statistical models, rule-based models (e.g., decision trees, Bayesian networks, etc.), and / or machine learning (ML) models or deep learning (DL) models. For example, an AI model may include a convolutional neural network (CNN) trained to detect lesions or anomalies within the breast image volume. One or more AI models may take the image volume as input and output an assessment of the breast. The assessment may include, for example, determining whether a lesion has been detected, estimating the severity and / or extent of the lesion, and the type of lesion (calcification cluster or mass).

[0033] In step 306, method 300 includes assigning a case score to a patient case based on integrating evaluations of each collected data from the patient's breast, and generating a corresponding image interpretation protocol and a minimum dataset of images to be reviewed according to that protocol. The case score can classify the evaluation into one of several categories or classifications. The case score can be an integer value (e.g., 1, 2). The case score can be assigned by a protocol generation system (e.g., protocol generation system 820 in Figure 8). Specifically, the case score can be generated by a protocol generation model of the protocol generation system (e.g., protocol generation model 830). The protocol generation model can take the evaluation of each collected data from the patient's breast as input and output a case score. In some embodiments, instead, the protocol generation system 820 can receive the case score as input from an external source. Based on the case score, an appropriate image interpretation protocol and minimum dataset can be selected.

[0034] In step 308, method 300 includes determining whether the case score is lower than a first threshold. For example, the case score may be zero and the first threshold may be 1. If the case score is less than the first threshold, it can be estimated with high confidence that no abnormalities were detected in the breast by one or more AI models, and method 300 proceeds to step 310. In step 310, method 300 does not select a minimum dataset and does not send patient cases to the radiologist's (e.g., radiologist 312) RWS, and method 300 terminates.

[0035] On the other hand, if the case score is greater than or equal to the first threshold, Method 300 proceeds to step 312. In step 312, Method 300 includes determining whether the case score is less than the second threshold. For example, the case score may be 1 and the first threshold is 2. If the case score is less than the second threshold, it can be estimated that the patient case contains findings that have been detected and diagnosed with high confidence by one or more AI models, and Method 300 proceeds to step 314. In step 314, Method 300 includes selecting a minimum dataset containing S2D images and an S2D protocol (e.g., an S2D-only protocol) to send to the radiologist's RWS, and Method 300 proceeds to step 322. An S2D-only protocol may instruct the radiologist to review findings from one or more AI models based on S2D images, but does not specify that the radiologist review DBT images. However, DBT images, DBT volumes, and / or other data and files of the patient case may be additionally sent to the radiologist for reference.

[0036] On the other hand, if the case score in step 312 is above the second threshold, it can be inferred that the patient case contains findings detected and diagnosed with low confidence and / or high severity by one or more AI models, in which case detailed review and confirmation by a human expert of the diagnosis of one or more AI models is desirable. For example, one or more malignant tumors may be detected from the breast by one or more AI models.

[0037] If the case score is greater than or equal to the second threshold in step 312, method 300 proceeds to step 316. In step 316, method 300 includes determining whether the case score is less than the third threshold. If the case score is less than the third threshold, method 300 proceeds to step 318. In step 318, method 300 includes selecting a minimum dataset containing S2D images and slabs, and the S2D+Slab protocol, to send to the radiologist's RWS. Method 300 then proceeds to step 322. The S2D+Slab protocol specifies that the radiologist will review findings in one or more AI models using S2D and slabs, but may not specify that the radiologist will review a full set of DBT images. However, DBT image volumes of patient cases, and / or other data and files may be additionally sent to the radiologist for reference.

[0038] On the other hand, in step 316, if the case score is above the third threshold, it can be inferred that the patient case contains findings detected and diagnosed with lower confidence and / or higher severity by one or more AI models, in which case detailed review and confirmation by a human expert of the diagnosis of one or more AI models is desirable. If the case score is above the third threshold in step 316, method 300 proceeds to step 320. In step 320, method 300 includes selecting an S2D+ cross-sectional+ slab protocol to be sent to a radiologist for review, in which case the minimum dataset is the full dataset of images of the patient case. The protocol may specify that the radiologist review the findings based on the full dataset.

[0039] In step 322, method 300 optionally includes applying a selective compression algorithm to selected case files (i.e., case files selected in step 314 for the S2D-only protocol, case files selected in step 318 for the S2D+slab protocol, or case files selected in step 320 for the S2D+section+slab protocol). A benefit of the selective compression algorithm is that different case files for transmission to the RWS and / or to the digital mammography system archive (e.g., archive 860) can be compressed in different ways, thereby improving transmission efficiency and / or reducing the time required to transmit case files to the RWS or archive and / or reducing the storage capacity of the RWS or archive. In particular, selected case files may include a minimum dataset to be interpreted (e.g., a dataset specified by the relevant interpretation protocol) and additional case files that can be used for reference purposes. For example, in a patient case where the case score is greater than a first threshold and less than a second threshold, the minimum dataset may include S2D images, and the additional case files may include a full dataset of DBT images. Radiologists receive S2D images, interpret them according to S2D-only protocols, and optionally refer to the full DBT image dataset during the examination of the S2D images.

[0040] To improve transmission efficiency and / or reduce the time it takes to send case files to the RWS and / or reduce the storage capacity of the RWS, the minimum dataset can be compressed using a first compression method, and additional case files or the remaining files of the full dataset can be compressed using a second compression method. For example, the first compression method can be a lossless compression method, and the second compression method can be a lossy compression method. By using lossless compression for files specified in the image interpretation protocol and lossy compression for additional files not specified in the image interpretation protocol, the total amount of data sent to the RWS can be reduced compared to another scenario where all case files are sent with lossless compression. The advantage of reducing the total amount of data sent to the RWS is that bandwidth consumption during transmission can be reduced, which makes the flow of information throughout the networked computing devices associated with the digital mammography system faster and more efficient, and further speeds up communication between devices. Furthermore, the amount of memory used to store the case files sent to the RWS can be reduced, which increases the amount of memory available for storing other data.

[0041] In step 324, method 300 includes sending the selected protocol and full dataset to the RWS and / or saving the selected protocol and full dataset to an archive. Then method 300 terminates.

[0042] Therefore, based on the output of one or more AI models, different interpretation protocols and different sets of images referenced by these protocols can be sent to the radiologist. If there are no findings and the patient's diagnosis is clear, the radiologist may not be asked to interpret the images. If there are findings, but the diagnosis is not complex and malignancy is not suspected, a first protocol can be sent to the radiologist specifying that only the S2D images be interpreted and other images not to be interpreted, allowing the radiologist to confirm the diagnosis of one or more AI models using the S2D images without relying on other breast images extracted from the DBT volume. If the findings indicate malignancy or a complex diagnosis, a second or third protocol can be sent to the radiologist, specifying that additional images or a full set of images be interpreted to confirm the findings are correct. Thus, in cases where a diagnosis can be made without reviewing a full set of DBT images, an advantage is that the number of images that the radiologist has to review can be reduced by using AI models. By reducing the number of images reviewed by radiologists in these cases, the time spent by radiologists during review and the amount of system resources consumed during image interpretation can be reduced.

[0043] In other words, one or more AI models are used to define the minimum dataset that a radiologist will use when making a diagnosis. This minimum dataset may include a single S2D image, a full set of DBT images, or a partial set of DBT images, where the DBT images include both slab and / or cross-sectional images. Furthermore, in various embodiments, any remaining image sets that are not part of the minimum dataset may be sent to the radiologist, providing an option to display these remaining image sets on a display device. To conserve resources and bandwidth, the minimum dataset may be stored using lossless compression, while the remaining image sets may be stored using lossy compression.

[0044] Another embodiment is shown in Figure 4, in which the protocol generation system is installed and operated on the radiologist's RWS. Figure 4 shows a method 400 for generating an image interpretation protocol based on a full dataset of patient cases received from a digital mammography system (such as the digital mammography system 100 in Figure 1). In contrast to method 300 in Figure 3, method 400 can be executed by the processor of the RWS (such as the processor 852 in Figure 8) based on instructions stored in the RWS's memory (such as memory 854 in Figure 8).

[0045] Method 400 begins in 402 and includes receiving a full dataset of patient cases from a digital mammography system. The full dataset may include patient data, a set of 2D and / or 3D image volumes reconstructed by the digital mammography system, and one or more outputs of one or more AI (anomaly detection) models.

[0046] In 403, Method 300 includes analyzing one or more images from a full dataset using one or more AI models (such as the anomaly detection model of the anomaly detection system 810 in Figure 8). As described in Method 300 above, AI models may include, for example, statistical models, rule-based models (e.g., decision trees, Bayesian neural networks, etc.), and / or machine learning (ML) models or deep learning (DL) models. For example, an AI model may include a convolutional neural network (CNN) trained to detect lesions or anomalies in breast image volumes. One or more AI models may take an image volume as input and output an assessment of the breast. The assessment may include, for example, determining whether a tumor has been detected and estimating the severity and / or extent of the tumor.

[0047] In step 404, method 400 includes assigning a case score to a patient case using the protocol generation model described in method 300. In step 406, if the case score is above a second threshold, it can be inferred that the patient case contains findings detected and diagnosed with low confidence and / or high severity by one or more AI models, in which case detailed review and confirmation by a human expert of the diagnosis of one or more AI models is desirable. Therefore, method 400 proceeds to step 407. In step 407, method 400 includes selecting an S2D+section+slab reading protocol for reading images from the full dataset, and in step 412, method 400 includes displaying images from the full dataset according to the S2D+section+slab reading protocol. Images from the full dataset can be displayed by a hanging protocol established within the reading protocol that specifies the display order of images from the minimum dataset. In other embodiments, as described in Method 300, it is possible to use a third threshold similarly to transmit the S2D+ slab interpretation protocol and minimum dataset, but this is not shown here for simplicity.

[0048] On the other hand, if the case score in step 406 is less than the second threshold, method 400 proceeds to step 408. In step 408, method 400 includes selecting an S2D-only interpretation protocol for displaying the S2D images. Similar to method 300, if the case score is less than the second threshold, it is presumed that the patient's case contains benign findings that have been detected and diagnosed with high confidence by one or more AI models, and the S2D images are considered sufficient to confirm the diagnosis of one or more AI models. In step 409, method 400 includes displaying the S2D images according to the S2D interpretation protocol.

[0049] If, in step 410, input is received from the RWS user (e.g., radiologist 112) requesting to view additional images from the full dataset, method 400 proceeds to step 412, where additional images from the full dataset are displayed. In this case, the hanging protocol is not established, and the user can select individual images from the additional images to display. If, in step 410, no user input is received to view additional images from the full dataset, method 400 proceeds to step 414. In step 414, method 400 includes continuing to display S2D images according to the S2D image protocol, and then method 400 terminates.

[0050] While embodiments of Method 400 consume significant bandwidth during the transmission of the full dataset from the digital mammography system to the RWS, an advantage of Method 400 over Method 300 is that processing the output of the AI ​​model used for anomaly detection and generating the interpretation protocol is performed using the RWS processor rather than the digital mammography system's processor. The digital mammography system can improve the overall computational efficiency of the system by generating multiple images to be sent to multiple RWSs and offloading processing to edge devices, and freeing up the processing power of the digital mammography system for other tasks. Another advantage of Method 400 over Method 300 is that radiologists can manually request the display of data that is not part of the minimum dataset. This is useful when the AI ​​system fails to properly identify the minimum dataset to be interpreted.

[0051] Another embodiment is shown in Figure 5. Figure 5 shows an alternative method 500 for generating an image interpretation protocol and selectively sending data files to the receiving RWS. Method 500 can be performed by the processor of the digital mammography system or by a computing device coupled to the digital mammography system. Method 500 begins in step 502, in which a DBT scan is performed, a volume is reconstructed from the DBT scan data, and an S2D image is generated. In step 504, Method 500 includes analyzing the volume using one or more AI models used to generate the S2D image, and in step 506, a case score is assigned based on the output of one or more AI models used to generate the S2D image for each DBT collection data for each patient case, and an image interpretation protocol is selected based on the case score. However, in contrast to Method 300, in step 508, the minimum dataset to be interpreted and the associated image interpretation protocol are encoded and described in the DICOM (Digital Imaging and Communications in Medicine) tag of the S2D image. In some cases, the output of one or more AI models can also be encoded and described in DICOM tags.

[0052] In step 510, the case file containing the S2D images is sent to the receiving RWS or stored in the digital mammography system's archive. Once the case file is received by the receiving RWS or other RWS, the interpretation protocol, minimum dataset, and / or the output of one or more AI models are extracted and decoded from the DICOM tag, and the relevant images can be displayed according to the extracted interpretation protocol.

[0053] By encoding the image interpretation protocol and describing it in a DICOM tag, the requesting RWS can receive the full dataset, extract the interpretation protocol, determine a specific image in the full dataset, and display it. In other words, the interpretation protocol is generated only once, and subsequent case file requests include the respective interpretation protocol for viewing the case file. In this way, the processing resources of the digital mammography system can be reduced, the efficiency of the digital mammography system can be increased, and the number of computational tasks that can be performed on the digital mammography system can be increased.

[0054] In some medical situations, combo acquisition can be performed, in which case 2D FFDM images are generated by the digital mammography system in addition to DBT images. Abnormal findings (such as calcifications) may be more easily identified in FFDM images than in DBT images. In another embodiment, FFDM image data can be used as additional input to a protocol generation model used to determine an appropriate interpretation protocol for interpreting either or both DBT and FFDM images.

[0055] Figure 6 shows a method 600 for generating an image interpretation protocol and sending the images included in the protocol to the RWS for viewing, the images including both DBT and FFDM images. Method 600 can be executed by the processor of the digital mammography system based on instructions stored in the memory of the digital mammography system and / or in the protocol generation system of the digital mammography system.

[0056] Method 600 begins in step 602. In step 602, Method 600 includes performing a DBT examination of the patient's breast, reconstructing an image volume from the data acquired during the DBT examination, and generating an S2D image from the image volume and / or DBT projection. In step 604, Method 600 includes performing an FFDM examination of the patient to generate a 2D FFDM image. In step 606, Method 600 includes analyzing either or both of the DBT volume and / or FFDM image using one or more AI models as described above. One or more AI models may accept DBT data as input, or FFDM data as input, or both DBT data and FFDM data as input. In various embodiments, an AI model (e.g., an AI model of an anomaly detection system 810) can be used to detect and identify calcifications in the FFDM image and / or DBT volume. In step 608, the output of the AI ​​model can be used to assign a case score to the patient case. In particular, in addition to the case scores mentioned above in Method 300, an additional case score can be created for situations in which calcification is detected in FFDM images.

[0057] In step 610, method 600 includes determining whether the case score is less than a first threshold, as described above in method 300. If the case score is less than the first threshold, in step 612, the minimum dataset is not selected and patient cases are not sent to the RWS for viewing. If the case score is greater than or equal to the first threshold, method 600 proceeds to step 614. Step 614 includes determining whether the case score is less than a second threshold. If the case score is less than the second threshold, method 600 proceeds to step 616.

[0058] In step 616, method 600 optionally includes integrating data from FFDM images and DBT volumes into S2D images to best represent calcifications and tumors. If data from FFDM and DBT images are integrated, in step 618, method 600 includes selecting the integrated S2D images as the minimum dataset and selecting an S2D interpretation protocol. On the other hand, if data from FFDM and DBT images are not integrated in step 616, in step 618, method 600 includes selecting FFDM images as the minimum dataset and selecting an FFDM interpretation protocol.

[0059] In step 614, if the case score is above the second threshold, method 600 proceeds to step 620. In step 620, method 600 includes determining whether the case score is below the third threshold. If the case score is below the third threshold, FFDM images alone may not be sufficient, and method 600 proceeds to step 622. In step 622, method 600 includes selecting a minimum dataset containing FFDM images and slabs, along with an FFDM + slab interpretation protocol.

[0060] If the case score is above the third threshold in step 620, it is inferred that calcification has been detected in the FFDM image, and method 600 proceeds to 624. In step 624, method 600 includes selecting a minimum dataset including FFDM images, sections, and slabs, and an FFDM+S2D+section+slab protocol for interpreting the full dataset of patient images together with the FFDM images. Preferably, the FFDM images are displayed preferentially over other images in the full dataset.

[0061] In step 626, the method includes sending the selected protocol and full dataset to the RWS for viewing and / or saving the selected protocol and full dataset to the digital mammography system archive. Then, method 600 ends.

[0062] In other words, in some cases, FFDM images may contain anatomical features (e.g., calcifications) that are not considered by some AI models used to detect abnormalities within the DBT volume. That is, some AI models may be designed for cancer detection but not for calcification detection. By including FFDM image data as input to the protocol generation model, it is possible to generate case scores that reflect the findings of the FFDM images in addition to the findings of the DBT volume. If calcifications are detected, it may be beneficial for radiologists to first confirm the calcifications in the FFDM images and then review the S2D images and other images in the full dataset. Furthermore, FFDM image data can be integrated with S2D images, in which case, in some cases, reviewing the patient case may be sufficient with an S2D-only protocol.

[0063] Another advantage of the systems and methods disclosed herein is that by using interpretation protocols selectively generated by a protocol generation system, image data can be prefetched rapidly from an archive (such as archive 860 in Figure 8), thereby improving prefetching efficiency. As described above, in some embodiments, DBT image data and / or FFDM image data are not sent to the RWS for review by a radiologist, and the DBT image data and / or FFDM image data can be sent to an archive (e.g., a database or a dedicated area in memory) for storage. Then, when a radiologist requests image data from the archive to review, the image data can be sent from the archive to the radiologist's RWS. Sending image data to the archive instead of the RWS reduces the overall bandwidth consumption while sending image data from the digital mammography system (e.g., scanner) to the RWS, and / or allows for efficient transmission of image data. For example, image data can be transferred from the digital mammography system to the archive via a first network, and image data can be transferred from the archive to the RWS via a second network. The first network may be configured to maximize data transfer efficiency between the digital mammography system and the archive, and the second network may be configured to maximize data transfer efficiency from the archive to multiple network-connected RWSs. The archive may be physically located near the digital mammography system.

[0064] To improve the efficiency of data transfer from an archive to multiple RWSs, a prefetch strategy can be used. In a prefetch strategy, image data related to a patient case is retrieved from the archive and stored in the RWS at a first time point, based on the radiologist's schedule for reviewing patient cases using the RWS. The radiologist then reviews the image data on the RWS at a second time point (e.g., a scheduled time point) after the first time point. Additionally or alternatively, a first portion of the image data (e.g., S2D images) may be retrieved at the first time point, and additional image data (e.g., the full dataset) may be retrieved at the second time point. Prefetching or retrieving image data in advance allows for more efficient management of data transfer to multiple RWSs. For example, data transfers can be scheduled during times of high bandwidth utilization, or data transfers can be performed at a slower speed without impacting the workflow.

[0065] Figure 7 shows a method 700 for prefetching DBT and / or FFDM images by utilizing case scores generated by a protocol generation system to improve data transfer efficiency between the archive and the receiving RWS, according to one embodiment. In some embodiments, the case scores may be received from an external source. Method 700 can be executed by the processor of the RWS (e.g., processor 852) based on instructions stored in the memory of the RWS (e.g., memory 854). In embodiments described herein, method 500 of Figure 5 can be executed as a prerequisite for executing method 700, in which case, before the DBT and / or FFDM images are stored in the archive, the protocol generation system on the digital mammography system or server determines the interpretation protocol or minimum dataset to be interpreted, the interpretation protocol or minimum dataset is encoded, described in the DICOM tag of the S2D image, and decoded by the RWS.

[0066] Method 700 begins with step 702. Step 702 includes determining whether a radiologist is scheduled to review patient cases (including DBT and / or FFDM images) using the RWS, or whether a case file for a patient in the digital mammography system is requested. In some cases, the radiologist may schedule the review and / or request or schedule the case file request by a workflow management application installed on the RWS. In other cases, the radiologist may schedule the review of patient cases by a healthcare facility or management team (e.g., by case management software running on the facility's server).

[0067] If, in step 702, it is determined that no schedule has been set for reviewing a case review on the RWS, or that no case file has been requested for a case review on the RWS, then method 700 proceeds to step 704. In step 704, method 700 includes waiting for a scheduled case review or a case file request. On the other hand, if, in step 702, it is determined that a schedule has been set for reviewing a case review on the RWS, or that a case file has been requested for a case review on the RWS, then method 700 proceeds to step 706.

[0068] In step 706, method 700 includes retrieving S2D images from an archive, and in step 708, method 700 includes extracting a reading protocol and case score for a patient case from the DICOM tags of the S2D image files. In some embodiments, the case score may be received as input from an external source. As previously stated, the reading protocol can specify a set of image files that the radiologist should prioritize reviewing. For example, the reading protocol could be an S2D reading protocol that specifies the reading of S2D images that are thought to contain sufficient data to confirm the patient's past diagnoses made by an AI algorithm. For example, if an FFDM image shows information related to a past diagnosis (e.g., calcification) that was not captured, or was not clearly or accurately captured, in the DBT volume, the reading protocol can specify that the FFDM image be read before other images. The reading protocol could be an S2D+section+slab protocol that specifies the reading of a full dataset of DBT images, and can further specify the order in which individual DBT images should be read.

[0069] Case scores generated by the protocol generation system or received from an external source may indicate the type or classification of the image interpretation protocol included in the DICOM tag. For example, the first case score may indicate an S2D protocol, the second case score may indicate an S2D + section + slab protocol, and the third case score may indicate that an FFDM image should be interpreted.

[0070] In step 710, method 700 includes determining whether the extracted case score is less than a first threshold. If the extracted case score is less than the first threshold, it can be inferred that the interpretation protocol is an S2D-only protocol. In this case, the S2D images are interpreted by the radiologist, but the other images in the full dataset are not. Method 700 proceeds to step 712, where the S2D protocol and S2D images are uploaded to the RWS memory with high priority and stored for later review by the radiologist, while the sections and / or slabs are uploaded with low priority. On the other hand, if in step 710 it is determined that the case score is greater than or equal to the first threshold, it is inferred that the full dataset can be interpreted by the radiologist, and method 700 proceeds to step 714. In step 714, method 700 includes retrieving the image data of the full dataset of DBT images (and / or DBT volumes) from the archive with equal priority, and storing the interpretation protocol and full dataset in the RWS memory for later review by the radiologist.

[0071] It should be understood that Method 700 represents a simplified procedure used to determine whether S2D / FFDM images should be prioritized over other DBT data based on the extracted image interpretation protocol. In other embodiments, Method 700 may include additional steps that allow different types or subsets of images to be retrieved at different priority levels based on different case scores. By retrieving images included in the minimum dataset with higher priority and images not included in the minimum dataset with lower priority, bandwidth availability for transmitting different types of data can be managed more efficiently, thereby improving the overall efficiency of the digital mammography system.

[0072] Figures 9 and 10 quantitatively illustrate an example of how the amount of image data transmitted from the digital mammography system to the RWS decreased as a result of applying the above system and method. Referring to Figure 9, an exemplary figure of a full dataset 900 of patient case images is shown, which represents the total amount of images that can typically be transferred to the RWS for display (prior art). The total amount of images includes S2D images 902, which are composite images containing four separate images (right mediolateral oblique image (RMLO) 910, left mediolateral oblique image (LMLO) 912, right craniocaudal image (RCC) 914, and left craniocaudal image (LCC) 916). In addition to the S2D images 902, 50 slice images each of RMLO, LMLO, RCC, and LCC 906 can be transferred based on a slice thickness of 1 mm and an average breast thickness of 5 cm. In addition to the slice image 906, 16 slab images 904 can be transferred. Each slab image is generated from 6 consecutive slices, and each slab image overlaps the previous slab image by 3 mm. As a result, the total number of images transferred in the full dataset (for example, in the S2D+section+slab-interpretation protocol) is 268 (for example, 4 images in S2D image 902, 64 images in slab image 904, and 200 images in slice image 906).

[0073] In contrast, Figure 10 shows an exemplary representation 1000 of multiple images that can be transferred to and displayed on the RWS when an image interpretation protocol is selectively generated by the protocol generation system described herein. The exemplary representation 1000 is based on a case score and image interpretation protocol assigned according to breast density, in which images of fatty (e.g., low-density) breasts are easier to diagnose as normal than images of dense breasts. That is, if the first breast is fatty, the first review of that first breast can be easily performed, and therefore the review can be performed by interpreting a small number of first images. On the other hand, if the second breast is dense, the second diagnosis of that second breast may be difficult, and therefore the review can be performed by interpreting a large number of second images. Breast density can be determined by the Breast Imaging Reporting and Data System (BIRADS) classification, where BIRADS classifications A and B correspond to fatty, low-density breasts. Classification C corresponds to dense breasts, and classification D corresponds to breasts with a higher density than classification C.

[0074] According to method 300 in Figure 3, for example, a first case score can be assigned to breast images with density within the BIRADS classification A and B range, thereby selecting the first interpretation protocol. A second case score can be assigned to breast images with density within the BIRADS classification C range, thereby selecting the second interpretation protocol. A third case score can be assigned to breast images with density within the BIRADS classification D range, thereby selecting the third interpretation protocol.

[0075] Method 300 in Figure 3 can be applied to a theoretical (e.g., typical) patient population in which 50% of patients are classified into BIRADS classification A and B, 40% into BIRADS classification C, and 10% into BIRADS classification D. Patients in BIRADS classification A and B can be assigned a first case score. The first case score indicates that a radiologist can perform a review of these patients using a protocol for interpreting S2D images 1002 (including four separate images: RMLO, LMLO, RCC, and LCC). Patients classified as BIRADS Class C may be assigned a second case score, which indicates that the radiologist can perform a review of these patients using a protocol for interpreting S2D images 1002 and 16 slab images 1004 (each slab image 1004 includes 4 individual RMLO, LMLO, RCC, and LCC images, totaling 68 images). Patients classified as BIRADS Class D may be assigned a third case score, which indicates that the radiologist can perform a review of these patients using a protocol for interpreting S2D images 1002 and 50 slice images 1006 (each slice image 1006 includes 4 individual RMLO, LMLO, RCC, and LCC images, totaling 204 images).

[0076] The total number of images included in each image interpretation protocol is weighted by the proportion of patients belonging to each classification, allowing us to calculate the relative reduction in the number of images sent to the RWS (the relative reduction in the number of images obtained by using the image interpretation protocols described above, rather than the conventional image interpretation protocols shown in Figure 9). For BIRADS classifications A and B, the first weighted total number of images sent is 4 × 0.5 (e.g., 50%) = 2 images. For BIRADS classification C, the second weighted total number of images sent is 68 × 0.4 (e.g., 40%) = 27.2 images. For BIRADS classification D, the third weighted total number of images sent is 204 × 0.1 (e.g., 10%) = 20.4 images. Adding up the first, second, and third weighted totals, the total weighted number of images sent to the RWS for interpretation is approximately 50 images (e.g., 2 + 27.2 + 20.4 = 49.6).

[0077] From this, it can be estimated that in an average patient case where the prior probability of the patient being classified as BIRADS A or B is 50%, the prior probability of being classified as BIRADS C is 40%, and the prior probability of being classified as BIRADS D is 10%, the average number of images generated by the protocol generation system described herein and designated to be sent to the RWS is 50. Compared to the 268 images typically sent as described in Figure 9, using the protocol generation system can reduce the amount of image data typically transmitted over the network for radiologist interpretation by 80%. Assuming statistical or historical cancer incidence rates in a given patient population, similar advantages can be obtained when case scores and interpretation protocols are generated based on cancer diagnoses made by AI models (e.g., the anomaly detection system 810 in Figure 5).

[0078] Accordingly, this specification discloses a protocol generation system that performs an evaluation of DBT image data and selectively provides a minimum dataset and interpretation protocol based on the score generated from the evaluation. In some cases, a minimum dataset and protocol consisting only of S2D may be sufficient. In other cases, an S2D + slab / section protocol can be provided when the reliability of automated diagnosis is low. By transmitting an appropriate amount of image data for review based on the score, the time radiologists spend interpreting DBT and composite 2D images can be reduced without compromising the quality of the review. Reducing the time required for image interpretation reduces the computational and memory resources of the computing system used for image interpretation. Furthermore, advantageously, the bandwidth consumed when transmitting images to radiologists can be reduced, allowing the bandwidth to be used for other imaging tasks.

[0079] The technical benefits of evaluating DBT image data and selectively determining the minimum dataset of images to be reviewed based on the score generated from that evaluation include reducing the amount of computational and memory resources used in the computing system used for image interpretation, and reducing the amount of bandwidth consumed when transmitting images to radiologists.

[0080] This disclosure also provides support for a protocol generation system for a digital mammography system. The protocol generation system includes a processor and memory for storing instructions, the instructions including memory that, when the instructions are executed, cause the processor to receive a patient's digital breast tomosynthesis (DBT) case image, the DBT case image comprising at least one set of projection images and / or image volumes of the patient's breast reconstructed from data acquired using a digital mammography system; generate a case score for the DBT case image by analyzing the case image; select a minimum dataset from the DBT case image to be interpreted based on the case score; and send the minimum dataset and / or an interpretation protocol specifying the minimum dataset to a radiology workstation (RWS) for interpretation, and / or save the minimum dataset to an archive. In a first embodiment of the system, the protocol generation system is installed in a radiology workstation (RWS) that is integrated with the digital mammography system, electronically coupled to the digital mammography system, or communicably coupled to the digital mammography system via a network. In a second embodiment of the system, which optionally includes the first embodiment, the received DBT case images are analyzed by one or more artificial intelligence (AI)-based anomaly detection systems, the case score is generated by a protocol generation model of the protocol generation system, the protocol generation model receives the output of one or more AI-based anomaly detection systems as input. In some embodiments, the case score may instead be generated using the output of one or more AI models. In some embodiments, an external source may instead receive the case score as input.In a third embodiment of the System, which optionally includes one or both of the first and second embodiments, the one or more AI-based anomaly detection systems include one of a risk assessment system, a breast density assessment system, an anomaly detection system used to generate a set of slabs, a set of sections, and / or a composite 2D (S2D) image from received DBT case images, and another anomaly detection system. In a fourth embodiment of the System, which optionally includes one or more embodiments from the first to third embodiments, or each of these embodiments, the output of the anomaly detection system used to generate a set of slabs and / or an S2D image includes at least one of the location of the anomaly in the received case image, the classification of the anomaly, an assessment of the severity of the anomaly, and a confidence score indicating the confidence with which the anomaly detection system assesses the severity of the anomaly. In a fifth embodiment of the system, which includes one or more embodiments from the first to fourth embodiments, or any selection of each embodiment, other instructions are stored in the memory, and when the other instructions are executed, the processor is instructed not to select a minimum dataset if the case score is less than a first threshold, to select a minimum dataset that includes the S2D images but does not include a set of slab images and a set of cross-sectional images if the case score is less than a second threshold, to select a minimum dataset that includes the S2D images and a set of slab images but does not include a set of cross-sectional images if the case score is less than a third threshold, and to select a minimum dataset that includes the S2D images, a set of slab images and a set of cross-sectional images if the case score is greater than the third threshold. In a sixth embodiment of the system, which includes one or more embodiments from the first to fifth embodiments, or any selection of each embodiment, full-field digital mammography (FFDM) images of the breast are input to the protocol generation model, and the case score is generated at least partially based on the FFDM images.In a seventh embodiment of the system, which includes one or more embodiments from the first to sixth embodiments or any selection of each embodiment, other instructions are stored in the memory, and when the other instructions are executed, the processor is instructed, depending on the classification of the FFDM images by the anomaly detection system, not to select a minimum dataset if the case score is less than a first threshold; to select a minimum dataset that includes the FFDM images but does not include the set of slab images and the set of cross-sectional images if the case score is also less than a second threshold; to select a minimum dataset that includes the FFDM images and the set of slab images but does not include the set of cross-sectional images if the case score is less than a third threshold; and to select a minimum dataset that includes the FFDM images, the set of slab images and the set of cross-sectional images if the case score is greater than the third threshold. In an eighth embodiment of the system, which includes one or more embodiments from the first to seventh embodiments or any selection of each embodiment, the minimum dataset and / or the output of the anomaly detection system are encoded and described in a Digital Image and Communication in Medicine (DICOM) tag of the S2D image. In a ninth embodiment of the system, which includes one or more embodiments from the first to eighth embodiments, or any selection of each embodiment, other instructions are stored in the memory, and when the other instructions are executed, the other instructions cause the processor to compress a minimum dataset using lossless compression, compress the remaining portion of the case images not included in the minimum dataset using lossy compression, and send both the minimum dataset and the remaining portion to the archive.

[0081] This disclosure also provides methodological support. The method includes receiving a digital breast tomosynthesis (DBT) case image of a patient's breast, wherein the DBT case image includes at least one set of projection images and / or image volumes of the breast; analyzing the received DBT case image using one or more artificial intelligence (AI)-based anomaly detection systems; generating a composite two-dimensional (S2D) image, a set of slab images, and a set of cross-sectional images from the DBT case image using one or more AI-based anomaly detection systems; assigning a case score to the DBT case image based on the output of one or more AI-based anomaly detection systems using an AI model; selecting a reading protocol for interpreting the DBT case image based on the assigned case score, wherein the reading protocol includes a minimum dataset for interpretation of the S2D image, the set of slab images, the set of cross-sectional images, and the DBT case image; and transmitting the reading protocol and the minimum dataset to a radiology workstation (RWS) for interpretation, and / or saving the minimum dataset to an archive. In a first embodiment of the present method, selecting an image interpretation protocol for interpreting the DBT case images based on the assigned case score, and transmitting the image interpretation protocol and the minimum dataset to the RWS, further includes not selecting the minimum dataset if the case score is less than a first threshold, selecting an S2D image interpretation protocol in which the minimum dataset includes S2D images but does not include a set of slab images or a set of cross-sectional images if the case score is less than a second threshold, selecting an S2D+slab image interpretation protocol in which the minimum dataset includes the S2D images and a set of slab images but does not include a set of cross-sectional images if the case score is less than a third threshold, and selecting an S2D+cross-sectional+slab image interpretation protocol in which the minimum dataset includes all of the S2D images, a set of slab images, and a set of cross-sectional images if the case score is greater than the third threshold.In a second embodiment of the Method, which optionally includes the first embodiment, the Method further includes encoding at least one of the image interpretation protocol, the minimum dataset, and the output of the anomaly detection system to describe the S2D image in a Medical Digital Image and Communication (DICOM) tag. In a third embodiment of the Method, which optionally includes one or both embodiments of the first and second embodiments, the Method further includes compressing the minimum dataset using lossless compression, compressing the remaining portion of the DBT image not included in the minimum dataset using lossy compression, and sending both the minimum dataset and the remaining portion to an archive. In a fourth embodiment of the Method, comprising one or more embodiments from the first to third embodiments, or any selection of each embodiment, the Method further includes receiving full-field digital mammography (FFDM) images of the patient's breast, using the AI ​​model to assign the case score to the DBT case images based on the output of the one or more AI-based anomaly detection systems and data from the FFDM images, integrating the data from the FFDM into the S2D images, including the FFDM images in the minimum dataset, and prioritizing the interpretation of the FFDM images in a selected interpretation protocol. In a fifth embodiment of the Method, comprising one or more embodiments from the first to fourth embodiments, or any selection of each embodiment, the AI ​​model receives the output of one or more AI-based anomaly detection systems and DBT case images as input. In a sixth embodiment of the Method, which includes one or more embodiments from the first to fifth embodiments, or any selection of each embodiment, the AI ​​model receives the output of an anomaly detection system as an additional input, the output including at least one of the following: the location of the anomaly in the received DBT case image, the classification of the anomaly, an assessment of the severity of the anomaly, and a confidence score indicating the confidence of the anomaly detection system in assessing the severity of the anomaly.

[0082] This disclosure also provides support for the method. The method includes receiving a digital breast tomosynthesis (DBT) study of a patient's breast from a digital mammography system, wherein the DBT study comprises at least one of synthetic two-dimensional (S2D) images, a set of slab images, a set of cross-sectional images, and an image volume; analyzing the DBT study using one or more artificial intelligence (AI)-based anomaly detection systems; assigning a case score to the DBT study using an AI model based on the output of one or more AI-based anomaly detection systems; selecting a reading protocol for interpreting the DBT study based on the assigned case score, wherein the reading protocol comprises a minimum dataset of the DBT study to be interpreted; and displaying the minimum dataset on a display device according to the reading protocol. In a first embodiment of the method, the reading protocol is extracted from a DICOM (Digital Image and Communication in Medicine) tag of the S2D image. In a second embodiment of the method, which optionally includes the first embodiment, the minimum dataset specified in the extracted image interpretation protocol is retrieved from the digital mammography system archive at a first time point, and the remaining images of the DBT study that are not included in the minimum dataset are retrieved at a second time point, which is later than the first time point.

[0083] In this specification, an element or step described in the singular and preceded by the words "a" or "an" should be understood not to exclude multiple such elements or steps unless the exclusion of multiple such elements or steps is explicitly stated. Furthermore, a reference to "one embodiment" of the invention is not intended to be construed as excluding the existence of additional embodiments that also incorporate the mentioned features. Furthermore, unless the opposite is explicitly stated, an embodiment "comprising, including, having" one or more elements having a particular characteristic may include additional such elements that do not possess that characteristic. The terms "including" and "in which" are used as plain language expressions for the terms "comprising" and "wherein," respectively. Terms such as "first," "second," and "third" are used merely as labels and are not intended to impose numerical requirements or specific positional orders on the subjects of those terms.

[0084] This specification discloses the present invention (including the best mode) using examples and enables those skilled in the art to practice the invention (e.g., to make and use an apparatus or system, and to perform a method incorporating it). The patentable scope of the present invention is defined by the claims and may include other examples that are conceivable to those skilled in the art. Such other examples are intended to be within the scope of the claims if they have structural elements that are indistinguishable from the language of the claims, or if they include equivalent structural elements that are substantially indistinguishable from the language of the claims. [Explanation of symbols]

[0085] 100 Digital Mammography Systems 101 Projection View 105 Projection 112 Radiologists 131 lesions 133 Compression Paddle 140 X-ray source 141 directions 143 directions 144 Arm 155 data collection lines 160 memory units 165 Archive Line 170 Display Units 180 User Interfaces 181 Mammography Technician 185 signal line 200 Workflows 202 Digital Mammography System 204 Model 206 S2D images 208 S2D images 210 DBT images 212 Radiologists 214 Display screen 802 Digital Mammography System 803 S2D generator 804 images 806 Image Analysis System 808 AI Module 820 Protocol Generation System 824 processors 826 Temporary memory 830 Protocol Generation Models 840 Image Interpretation Protocol 842 minimum datasets 850 RWS 852 processors 854 memory 856 displays 860 Archives 900 full datasets 902 S2D image 904 Slab image 906 sliced ​​images 910 Image (Right-sided esotropia / exophoria image (RMLO)) 912 Left-sided endoscopic-external oblique view (LMLO) 914 Right temporo-caudal image (RCC) 916 Left temporo-caudal image (LCC) 1002 S2D images 1004 Slab image 1006 slice images

Claims

1. A protocol generation system for a digital mammography system, wherein the protocol generation system is A processor and a memory for storing instructions, wherein when an instruction is executed, the processor... Receiving a patient's digital breast tomosynthesis (DBT) case image, wherein the DBT case image includes at least one set of projection images and / or image volumes of the patient's breast reconstructed from data acquired using a digital mammography system. To generate or receive case scores as input for the aforementioned DBT case images, Based on the aforementioned case score, select the minimum dataset from the DBT case images that should be interpreted, and Memory that, along with the image interpretation protocol specifying the minimum dataset, transmits the minimum dataset and / or the full dataset to a radiology workstation (RWS) for image interpretation, and / or saves the minimum dataset to an archive. A protocol generation system that includes this.

2. The protocol generation system according to claim 1, wherein the protocol generation system is installed in a radiology workstation (RWS) that is integrated with the digital mammography system, electronically coupled to the digital mammography system, or communicatively coupled to the digital mammography system via a network.

3. The protocol generation system according to claim 1, wherein received DBT case images are analyzed by one or more artificial intelligence (AI) models, and the case score is generated using the output from the one or more AI models.

4. The aforementioned one or more AI models are Risk assessment system, Negative triage system, Breast density evaluation system, An anomaly detection system used to generate a set of cross-sections, a set of slabs, and / or a composite 2D (S2D) image from received DBT case images, and Another anomaly detection system The protocol generation system according to claim 3, comprising one of the following.

5. The output of the AI ​​model used to generate a set of cross-sections, slabs, and / or S2D images is Location of abnormalities in the received case images, Classification of the aforementioned abnormalities, Evaluation of the severity of the aforementioned abnormality, and A confidence score indicating the confidence level of the anomaly detection system when evaluating the severity or presence of the anomaly. The protocol generation system according to claim 4, comprising at least one of the following.

6. Other instructions are stored in the memory, and when the other instructions are executed, the processor receives the instructions. If the case score value falls within the first range, the minimum dataset is not selected. If the value of the case score falls within the second range, select the minimum dataset that includes the S2D images but does not include the set of slab images or the set of cross-sectional images. If the case score value falls within the third range, select the minimum dataset that includes the S2D image and slab image set but does not include the cross-sectional image set. If the case score value falls within the fourth range, the system prompts the user to select the minimum dataset including the S2D image, slab image set, and cross-sectional image set. The protocol generation system according to claim 1.

7. The protocol generation system according to claim 3, wherein full-field digital mammography (FFDM) images of the breast are input to the AI ​​model, and the case score is generated at least partially based on the FFDM images.

8. Other instructions are stored in the memory, and when the other instructions are executed, the processor is ordered to perform the following actions according to the classification of the FFDM image by the anomaly detection system: If the case score value falls within the first range, the minimum dataset is not selected. If the value of the case score falls within the second range, select the minimum dataset that includes the FFDM images but does not include the set of slab images or the set of cross-sectional images. If the value of the case score falls within the third range, select the minimum dataset that includes the set of FFDM images and slab images but does not include the set of cross-sectional images. If the case score value falls within the fourth range, the system prompts the user to select the minimum dataset including the FFDM images, slab images, and cross-sectional images. The protocol generation system according to claim 7.

9. The protocol generation system according to claim 5, wherein the output of the image interpretation protocol and / or anomaly detection system is encoded and described in a Digital Image and Communication (DICOM) tag for the S2D image.

10. Other instructions are stored in the memory, and when the other instructions are executed, the processor receives the instructions. Use lossless compression to compress the smallest dataset. Lossy compression is used to compress the remaining portion of the case images that is not included in the minimum dataset. This process involves sending both the minimum dataset and the remaining portion to the archive. The protocol generation system according to claim 1.

11. Receiving a digital breast tomosynthesis (DBT) case image of a patient's breast, wherein the DBT case image includes at least one set of projection images and / or image volumes of the breast. Analyzing received DBT case images using one or more artificial intelligence (AI)-based anomaly detection systems. Using one or more AI-based anomaly detection systems, generate a composite two-dimensional (S2D) image, a set of slab images, or a set of cross-sectional images from the DBT case images. Assigning a case score to the DBT case image based on the output of the one or more AI-based anomaly detection systems, Based on the assigned case score, select an image interpretation protocol and a minimum dataset for interpreting the DBT case images, wherein the image interpretation protocol includes, at a minimum, a reference to the images in the minimum dataset, and Along with the image interpretation protocol, transmit the minimum dataset and / or full dataset to the radiology workstation (RWS) for image interpretation, and / or save the minimum dataset to an archive. Methods that include...

12. Selecting an image interpretation protocol for interpreting the DBT case images based on the assigned case score, and transmitting the image interpretation protocol and the minimum dataset to the RWS, further, If the case score value falls within the first range, the minimum dataset is not selected. If the value of the case score falls within the second range, select an S2D image interpretation protocol in which the minimum dataset includes S2D images but does not include a set of slab images or a set of cross-sectional images. If the value of the case score is within the third range, select an S2D + slab image interpretation protocol in which the minimum dataset includes the set of S2D images and slab images but does not include the set of cross-sectional images, and If the case score value falls within the fourth range, select the S2D + cross-sectional + slab image interpretation protocol, which includes all of the S2D images, slab images, and cross-sectional images in the minimum dataset. The method of claim 11, including the method of claim 11.

13. The method according to claim 11, further comprising encoding at least one of the image interpretation protocol and the output of the anomaly detection system and describing the S2D image in a medical digital image and communication (DICOM) tag.

14. Compressing the minimum dataset using lossless compression, Using lossy compression, compress the remaining portion of the DBT image that is not included in the minimum dataset, and Send both the minimum dataset and the remaining portion to the archive. The method according to claim 12, further comprising:

15. To receive full-field digital mammography (FFDM) images of the patient's breast, Assigning the case score to the DBT case image based on the output of the one or more AI-based anomaly detection systems that have received the DBT case image and / or the FFDM image as input. To integrate the data from the FFDM into the S2D image, The FFDM image is included in the minimum dataset, and Prioritize the interpretation of the FFDM image in the selected image interpretation protocol. The method according to claim 11, further comprising:

16. The AI-based anomaly detection system is Location of abnormalities in received DBT case images, Classification of the aforementioned abnormalities, Evaluation of the severity of the aforementioned abnormality, A confidence score indicates the reliability of the anomaly detection system when evaluating the severity and / or presence of anomalies. The method according to claim 15, comprising at least one of the following.

17. Receiving a digital breast tomosynthesis (DBT) study of a patient's breast from a digital mammography system, wherein the DBT study includes at least one of a composite two-dimensional (S2D) image, a set of slab images, a set of cross-sectional images, and an image volume. Analyzing the DBT study using one or more artificial intelligence (AI)-based anomaly detection systems, Using an AI model, assign case scores to the DBT study based on the output of one or more AI-based anomaly detection systems. Selecting an image interpretation protocol for interpreting the DBT study based on the assigned case score, wherein the image interpretation protocol includes, at a minimum, a reference to images of the smallest dataset to be interpreted, and Display the minimum dataset on the display device in accordance with the aforementioned image interpretation protocol. Methods that include...

18. The method of claim 17, wherein the image interpretation protocol is extracted from a Digital Image and Communication (DICOM) tag in a medical S2D ​​image.

19. The method according to claim 18, wherein the minimum dataset specified in the extracted image interpretation protocol is retrieved from the digital mammography system archive at a first time point, and the remaining images of the DBT study that are not included in the minimum dataset are retrieved at a second time point later than the first time point.

Images