An ultrasound-assisted diagnosis system for lower extremity deep vein thrombosis based on video dynamic operators

Abstract

The present invention belongs to the technical field of intelligent auxiliary diagnosis of lower extremity venous ultrasound, and provides a lower extremity venous thrombosis ultrasound auxiliary diagnosis system based on a video dynamic operator. The lower extremity ultrasound video obtained is segmented, and the segmented ultrasound video is used for lower extremity venous thrombosis ultrasound diagnosis based on the video dynamic operator model. Specifically, first, the motion direction of each frame in the segmented ultrasound video is extracted as a motion operator feature. Then, the difference between the current frame and the previous frame in the ultrasound video is extracted, and the similarity index is calculated as another motion operator feature. Finally, the learning clustering algorithm is used to learn the difference between having a thrombus and not having a thrombus to obtain the diagnosis result of having or not having a thrombus. Based on the video dynamic operator, early screening can be carried out for symptomatic patients who have not been found to have deep vein thrombosis, improving the diagnosis efficiency, reducing the waiting time for patients to be referred, and reducing the workload of doctors.

Description

Technical Field

[0001] This invention belongs to the field of intelligent ultrasound-assisted diagnosis technology for lower extremity veins, and particularly relates to an ultrasound-assisted diagnosis system for lower extremity vein thrombosis based on video dynamic operators. Background Technology

[0002] Deep vein thrombosis (DVT) of the lower extremities is a common peripheral vascular disease. DVT-induced venous valve insufficiency and subsequent pulmonary embolism pose a significant risk to patients' work capacity and life safety. DVT has always been a major concern in clinical practice, and leg compression ultrasound is the gold standard for diagnosing the presence of thrombosis. When the ultrasound probe presses on a vein, if the vein is completely closed, it indicates the absence of a blood vessel; if the vein is partially or fully open, it indicates the presence of a thrombus.

[0003] The inventors discovered that many patients who may have symptoms do not have deep vein thrombosis when diagnosing lower extremity deep vein thrombosis. Therefore, when using leg compression ultrasound to diagnose the presence of thrombosis, it is not possible to diagnose and screen patients with symptoms but no deep vein thrombosis in the early stages, resulting in long waiting times for referrals and placing a heavy clinical burden on specialists. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes an ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators. Utilizing segmented ultrasound video, the system performs ultrasound diagnosis of lower extremity venous thrombosis based on a video dynamic operator model, enabling early screening and assisted diagnosis of patients. This significantly reduces the workload of doctors and improves diagnostic efficiency.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: In a first aspect, the present invention provides an ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators, comprising a video segmentation module and a processing module connected to each other; The video segmentation module is configured to segment the lower limb ultrasound video; The processing module is configured to: perform ultrasound diagnosis of lower extremity venous thrombosis using segmented ultrasound video based on a video dynamic operator model; specifically, firstly, the motion direction of each frame in the segmented ultrasound video is extracted as a motion operator feature; then, the difference between the current frame and the previous frame in the ultrasound video is extracted, and a similarity index is calculated as another motion operator feature; finally, a learning clustering algorithm is used to learn the difference between thrombosis and non-thrombosis to obtain the diagnostic result of thrombosis.

[0006] Furthermore, it also includes an acquisition module, configured to acquire ultrasound video of the lower limb to be tested before the processing module performs its processing.

[0007] Furthermore, a continuous ultrasound video of the entire process of the lower limb being tested changing from a state without pressure to a state under pressure, and then to a state where pressure is lifted.

[0008] Furthermore, the ultrasound video includes long-axis video and short-axis video.

[0009] Furthermore, in ultrasound video segmentation, labels are created only for keyframes, while pseudo-labels are constructed for the remaining frames based on temporal information and similarity between frames.

[0010] Furthermore, the similarity between frames is calculated. If the similarity between the segmentation result of the current frame and the segmentation result of the previous frame is greater than a preset value, then the current frame is defined as a key frame.

[0011] Furthermore, based on optical flow motion, the motion direction of each frame in the ultrasound video is extracted as a motion operator feature; the difference between the current frame and the previous frame in the ultrasound video is extracted, and the similarity index is calculated as another motion operator feature; finally, the two operator features are subjected to dimensionality reduction processing.

[0012] Furthermore, the ultrasound video is segmented into vascular contours and arterial contours, with the arterial contours used for auxiliary location determination.

[0013] Furthermore, the ultrasound video is segmented to obtain the vein contour; it is then determined whether the vein contour is in a closed state. If it is, the diagnosis is no thrombosis; otherwise, the diagnosis is thrombosis.

[0014] Furthermore, the outlines of veins with thrombosis are classified; veins with a partially open outline are classified as partially blocked by thrombosis, while veins with a fully open outline are classified as completely blocked by thrombosis.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention segments acquired lower limb ultrasound videos and uses the segmented ultrasound videos to perform ultrasound diagnosis of lower limb deep vein thrombosis based on a video dynamic operator model. Specifically, firstly, the motion direction of each frame in the segmented ultrasound video is extracted as a motion operator feature; then, the difference between the current frame and the previous frame in the ultrasound video is extracted, and a similarity index is calculated as another motion operator feature; finally, a learning clustering algorithm is used to learn the difference between thrombosis and non-thrombosis to obtain a diagnostic result of thrombosis presence or absence. Based on the video dynamic operator, early screening can be performed on symptomatic patients who have not yet been diagnosed with deep vein thrombosis, improving diagnostic efficiency, reducing patient waiting time for referral, and alleviating the workload of doctors. 2. This invention uses a segmentation model based on ultrasound video streams to achieve thrombus identification using video dynamic operators, solving the problems of insufficient labels and small sample sizes in lower limb veins, further improving the accuracy of the model, and providing a new end-to-end approach to thrombus detection based on ultrasound video. Attached Figure Description

[0016] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0017] Figure 1 This is a flowchart of Embodiment 1 of the present invention; Figure 2 This is a model framework diagram based on video dynamic operators in Embodiment 1 of the present invention. Detailed Implementation

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0019] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0020] Example 1: Leg compression ultrasound is the gold standard for diagnosing thrombosis. The principle is that when the ultrasound probe presses on a vein, if the vein is completely closed, it indicates the absence of blood vessels; if the vein is partially or fully open, it indicates the presence of a thrombus. However, in reality, many patients who may have symptoms do not have deep vein thrombosis (DVT). Therefore, when the ultrasound probe presses on a vein, the vein may not appear partially or fully open, or the partial or full opening may be indistinct. In such cases, diagnosis and early screening of symptomatic patients without DVT cannot be performed, leading to long waiting times for referrals and placing a significant clinical burden on specialists.

[0021] To address the aforementioned issues, this implementation provides an ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators. By acquiring and processing ultrasound images of lower extremity veins, it automatically identifies the presence of thrombotic diseases and determines the severity of thrombosis, thus assisting doctors in diagnosis, reducing their workload, and improving diagnostic efficiency. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators includes interconnected video segmentation and processing modules. The video segmentation module is configured to segment the lower limb ultrasound video; The processing module is configured to: perform ultrasound diagnosis of lower extremity venous thrombosis using segmented ultrasound video based on a video dynamic operator model; specifically, firstly, the motion direction of each frame in the segmented ultrasound video is extracted as a motion operator feature; then, the difference between the current frame and the previous frame in the ultrasound video is extracted, and a similarity index is calculated as another motion operator feature; finally, a learning clustering algorithm is used to learn the difference between thrombosis and non-thrombosis to obtain the diagnostic result of thrombosis.

[0022] Video dynamic operators can be used to screen symptomatic patients who have not yet been diagnosed with deep vein thrombosis, improving diagnostic efficiency, reducing patient waiting time for referral, and alleviating the workload of doctors.

[0023] To ensure complete ultrasound video and improve diagnostic accuracy, the ultrasound video includes long-axis and short-axis videos. Long-axis videos refer to a sequence of ultrasound images acquired with the sonographic section located at the maximum longitudinal section of the vein, while short-axis videos refer to a sequence of ultrasound images acquired with the sonographic section perpendicular to the long axis of the vein. Short-axis videos observe the compressibility of veins. For normal veins, the lumen can be compressed to complete closure under a small amount of external force generated by probe pressure; this is considered the gold standard for diagnosing thrombosis. Long-axis sections are prone to misinterpretation due to probe lateral movement or changes in vessel position, causing the sonographic image to deviate from the maximum longitudinal section and appear as if the vessel is compressed. Long-axis videos focus on observing the location, size, echogenicity, and blood flow signals of thrombi, playing an auxiliary role in intelligent thrombus identification.

[0024] Optionally, the acquisition module can be understood as a software module, circuit module, or ultrasound equipment module. When using ultrasound equipment, it can include, but is not limited to, ultrasound acquisition instruments, handheld ultrasound devices, and 5G remote ultrasound acquisition devices. Unlike the traditional ultrasound equipment product form of a main unit plus a probe, the main unit is miniaturized into a very small circuit board built into the probe. One probe is equivalent to an ultrasound machine, which can be displayed using a mobile phone or tablet with an ultrasound APP software installed. Image and other information are transmitted to the mobile phone or tablet via the probe's built-in Wi-Fi, and ultrasound video is transmitted to the processing module via the probe's built-in Wi-Fi. It is understood that the acquisition module and the processing module can be connected via wired connection, wireless connection, or other connection methods that can realize the connection between modules.

[0025] To achieve dynamic and continuous diagnosis of the patient's lower limbs and improve diagnostic accuracy, this embodiment uses continuously acquired ultrasound video. It captures a continuous ultrasound video stream of the entire process of the lower limb changing from a state without pressure to a state under pressure, and then to the end of pressure. Specifically, by pressing the ultrasound probe, a complete continuous ultrasound video stream is acquired from the state without pressure to the state under pressure, and then to the end of pressure. The acquisition of this continuous ultrasound video stream ensures that the ultrasound video contains complete information about veins and arteries, resulting in higher accuracy in thrombosis diagnosis after segmentation of the lower limb venous and arterial vessel contours.

[0026] The processing module can be understood as a software module, circuit module, or processor, etc., and may include a video segmentation module, a thrombus detection module, and a thrombus classification module.

[0027] The video segmentation module can be used to segment the received ultrasound video in real time to obtain the outlines of lower limb veins and arteries. The segmentation of the arterial outline is used to assist in the location determination, while the vein outline plays the most important role in the next step of thrombus identification.

[0028] The video segmentation module includes a unit to improve segmentation accuracy and a unit to reduce video redundancy computation; optional: The segmentation accuracy improvement unit is used to improve the segmentation accuracy of ultrasound video by using the temporal information of ultrasound video to obtain features with stronger semantic information consistency for segmentation.

[0029] The video redundancy reduction unit is used to reduce the computational load of the segmentation model by utilizing the similarity between frames to reduce the computational load of the segmentation model and improve the running speed and throughput of the model.

[0030] Optionally, the ultrasound video segmentation model primarily employs a semi-supervised keyframe adaptive model. Specifically, semi-supervised means that in ultrasound video segmentation, only a subset of keyframes are labeled, while pseudo-labels are constructed for the remaining frames based on temporal information and frame similarity. The keyframe adaptive model calculates the similarity between frames; if the similarity between the current frame's segmentation result and the previous frame's segmentation result is too high (e.g., greater than a preset value), it indicates a significant change in the scene, and the current frame is then considered a keyframe.

[0031] The thrombosis detection module is used to determine whether there is a thrombosis in the veins of the lower limbs based on a video-based dynamic operator model. If a thrombosis is found, it is sent to the next step of the thrombosis classification model.

[0032] Optionally, video-based dynamic operator models such as Figure 2As shown, firstly, based on optical flow motion, the motion direction of each frame in the ultrasound video is extracted as a motion operator feature; then, the difference between the current frame and the previous frame in the ultrasound video is extracted, and a similarity index is calculated as another motion operator feature; then, the two operator features are input into the manifold space for dimensionality reduction processing, and a manifold learning clustering algorithm is used to learn the difference between thrombosis and non-thrombosis, finally outputting the result of thrombosis presence or absence.

[0033] The thrombus classification module is used to classify ultrasound videos containing thrombi, mainly into two categories: partial occlusion and complete occlusion. Specifically, it determines whether the vein is partially open or fully open based on the outline of the venous thrombus. If it is partially open, it outputs "partial occlusion"; if it is fully open, it outputs "complete occlusion".

[0034] The acquired ultrasound video is segmented to obtain the venous vessel contour. Based on the acquired lower limb ultrasound video, the video is segmented to obtain the venous vessel contour, and then the presence of venous closure in the venous vessel contour is determined to diagnose lower limb deep vein thrombosis. It can efficiently and accurately determine whether the vein is partially or fully open, enabling early screening of symptomatic patients who do not show signs of deep vein thrombosis, improving diagnostic efficiency and reducing patient waiting time for referral.

[0035] Optionally, thrombosis classification networks may include, but are not limited to, mainstream classification networks such as VGG (Visual Geometry Group), GoogleNet, and LeNet.

[0036] The system also includes a display module connected to the processor, used to display the segmentation results of lower limb veins, the identification of venous thrombosis, and the classification results in real time.

[0037] This embodiment utilizes a semi-supervised video method to screen for lower extremity venous thrombosis, which improves the accuracy of identification, reduces the burden on doctors, and increases diagnostic efficiency.

[0038] Example 2: This embodiment provides an ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators, characterized in that it includes an ultrasound device and a processor connected to the ultrasound device; The ultrasound device is configured to acquire ultrasound video of the lower limb to be tested before the processing module is executed. The processor is configured to: segment the ultrasound video to obtain a vein contour; determine whether the vein contour is in a closed state; if so, diagnose as no thrombosis; otherwise, diagnose as thrombosis.

[0039] The ultrasound device and processor in this embodiment have all the features of the acquisition module and processing module in Embodiment 1, and will not be described again here.

[0040] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A lower extremity venous thrombosis ultrasound-assisted diagnostic system based on video dynamic operators, characterized in that, This includes interconnected video segmentation and processing modules; The video segmentation module is configured to segment the lower limb ultrasound video; The processing module is configured to: perform ultrasound diagnosis of lower extremity venous thrombosis using segmented ultrasound video based on a video dynamic operator model; specifically, firstly, the motion direction of each frame in the segmented ultrasound video is extracted as a motion operator feature; then, the difference between the current frame and the previous frame in the ultrasound video is extracted, and a similarity index is calculated as another motion operator feature; finally, a learning clustering algorithm is used to learn the difference between thrombosis and non-thrombosis to obtain the diagnostic result of thrombosis.

2. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators as described in claim 1, characterized in that, It also includes an acquisition module, configured to acquire ultrasound video of the lower limb to be tested before the processing module performs its processing.

3. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators as described in claim 1, characterized in that, A continuous ultrasound video of the lower limb being tested was collected, showing the entire process from no pressure to pressure, and then to the end of pressure.

4. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators as described in claim 1, characterized in that, The ultrasound video includes long-axis video and short-axis video.

5. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators as described in claim 1, characterized in that, In ultrasound video segmentation, labels are created only for keyframes, and pseudo-labels are constructed for the remaining frames based on the temporal information and similarity between frames.

6. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators as described in claim 5, characterized in that, Calculate the similarity between frames. If the similarity between the segmentation result of the current frame and the segmentation result of the previous frame is greater than a preset value, then the current frame is defined as a key frame.

7. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators as described in claim 6, characterized in that, Based on optical flow motion, the motion direction of each frame in the ultrasound video is extracted as a motion operator feature; the difference between the current frame and the previous frame in the ultrasound video is extracted, and the similarity index is calculated as another motion operator feature; finally, the two operator features are subjected to dimensionality reduction processing.

8. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators as described in claim 1, characterized in that, The ultrasound video is segmented into vascular contours and arterial contours, with the arterial contours used to assist in determining the location.

9. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators as described in claim 1, characterized in that, The ultrasound video is segmented to obtain the outline of the vein; it is then determined whether the vein outline shows a closed state. If it does, the diagnosis is no thrombosis; otherwise, the diagnosis is thrombosis.

10. The ultrasound-assisted diagnostic system for lower extremity venous thrombosis based on video dynamic operators as described in claim 9, characterized in that, The outlines of veins containing thrombi are classified; veins with a partially open outline are classified as partially blocked by thrombi, while veins with a fully open outline are classified as completely blocked by thrombi.

Images