Thyroid nodule detection system
By employing frame-by-frame detection and a multi-level false positive filtering mechanism, the problem of high false positive rate in thyroid nodule detection systems has been solved, thus reducing the false positive rate and improving diagnostic accuracy and screening efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 聆数医疗科技(苏州)有限公司
- Filing Date
- 2026-03-04
- Publication Date
- 2026-07-21
AI Technical Summary
Existing thyroid nodule detection systems have a high false positive rate in ultrasound images, which increases the workload of doctors in verification, especially in screening scenarios, and reduces screening efficiency.
A frame-by-frame detection module is used to detect target objects in thyroid ultrasound videos. Combined with a trajectory determination module and a video quality control module, a multi-level false positive filtering mechanism, including anatomical, continuity and morphological filtering mechanisms, is used to reduce the false positive rate.
To reduce the false positive rate, minimize unnecessary doctor reviews and patient anxiety, improve diagnostic accuracy and screening efficiency, and ensure diagnostic consistency.
Smart Images
Figure CN122434813A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of image processing technology and medical device technology, and in particular to a thyroid nodule detection system. Background Technology
[0002] Due to the high noise and artifacts in ultrasound images, existing thyroid nodule detection systems generally have a high false positive rate. This problem is particularly severe in screening scenarios, significantly increasing the workload for doctors and reducing screening efficiency. Therefore, an effective solution is urgently needed to address these issues. Summary of the Invention
[0003] To address the above problems, the present invention provides a thyroid nodule detection system.
[0004] This invention provides a thyroid nodule detection system, comprising: The frame-by-frame detection module is configured to perform frame-by-frame detection of target objects contained in a thyroid ultrasound video, and obtain detection data of each target object in each video frame. The detection data includes a segmentation mask, a detection box, and a detection confidence score. Each target object includes a carotid artery, a thyroid gland, and a thyroid nodule. The trajectory determination module is configured to determine each tracking trajectory of each target object based on each detection box and each detection confidence level; The video quality control module is configured to detect whether the thyroid ultrasound video is qualified based on each tracking trajectory of the carotid artery and / or each tracking trajectory of the thyroid gland; The nodule classification module is configured to, under the condition that the quality control of the thyroid ultrasound video is qualified, determine the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video by adopting a multi-level false positive filtering mechanism based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule. The multi-level false positive filtering mechanism includes an anatomical filtering mechanism, a continuity filtering mechanism, and a morphological filtering mechanism.
[0005] According to the thyroid nodule detection system provided by the present invention, the video quality control module is specifically configured as follows: The tracking trajectories of the thyroid gland are arranged in descending order of length to obtain the first trajectory sequence; If a first target tracking trajectory exists in the first N tracking trajectories of the first trajectory sequence, the thyroid ultrasound video quality control is deemed qualified. If the first target tracking trajectory is not found in the first N tracking trajectories of the first trajectory sequence, the thyroid ultrasound video quality control is determined to be unqualified. Wherein, the first target tracking trajectory is a tracking trajectory that meets the first set quality control conditions; The first set quality control conditions include: the trajectory length of the tracking trajectory is greater than a length threshold, the average confidence value of the tracking trajectory is greater than a trajectory confidence threshold, and the number of video frames in which the detection box of the thyroid gland in the tracking trajectory is located in the center of the video frame is greater than a number threshold.
[0006] According to a thyroid nodule detection system provided by the present invention, the nodule classification module is specifically configured as follows: Based on the aforementioned anatomical filtering mechanism, the aforementioned continuity filtering mechanism, and the aforementioned morphological filtering mechanism, criteria for identifying nodules, suspected nodules, and false positive nodules are established. Based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule, the target parameters of the thyroid nodule are determined; The target parameters are compared with the criteria for confirmed nodules, the criteria for suspected nodules, and the criteria for false positive nodules, respectively, to obtain the detection results of the target thyroid nodules corresponding to the thyroid ultrasound video.
[0007] According to the thyroid nodule detection system provided by the present invention, the criteria for determining nodules include: The number of consecutive frames in which the thyroid nodules appear is greater than or equal to the first frame number threshold; The mean confidence level of the detection of the thyroid nodules is greater than or equal to a first mean threshold. The average size of the thyroid nodules is greater than or equal to the size threshold; The thyroid nodule is located inside the thyroid gland; Furthermore, the thyroid nodule and the carotid artery do not overlap.
[0008] According to the thyroid nodule detection system provided by the present invention, the criteria for suspected nodules include: The thyroid nodule is located inside the thyroid gland; The thyroid nodule and the carotid artery do not overlap. At least one of the following parameters—the number of consecutive frames of nodules in the thyroid nodule, the mean of the detection confidence of the thyroid nodule, and the average size of the thyroid nodule—is less than the threshold corresponding to the parameter, and the difference between the threshold corresponding to the parameter and the parameter is less than the set value corresponding to the parameter.
[0009] According to the thyroid nodule detection system provided by the present invention, the false positive nodule criteria include at least one of the following: The overlap between the thyroid nodule and the carotid artery is greater than the overlap threshold; The thyroid nodule is located outside the thyroid gland; The number of consecutive frames of the thyroid nodules is less than a second frame number threshold, and / or the mean of the detection confidence of the thyroid nodules is less than a second mean threshold.
[0010] According to the present invention, a thyroid nodule detection system is provided, wherein the anatomical filtering mechanism characterizes that the thyroid nodule and the carotid artery do not overlap and that the thyroid nodule is located inside the thyroid gland; The continuous filtering mechanism is characterized by the number of consecutive frames of the thyroid nodules being greater than or equal to a first frame number threshold and the mean of the detection confidence of the thyroid nodules being greater than or equal to a first mean threshold. The morphological filtering mechanism characterizes the average size of the thyroid nodules as being greater than or equal to a size threshold.
[0011] A thyroid nodule detection system according to the present invention further includes: The nodule measurement and statistics module is configured to determine the minimum bounding rectangle of each segmentation mask of the thyroid nodule; determine the size of the thyroid nodule based on the width and height of each minimum bounding rectangle; count the number of thyroid nodules in the thyroid ultrasound video, and classify the thyroid nodules according to their size.
[0012] A thyroid nodule detection system according to the present invention further includes: The report frame selection module is configured to, when the thyroid nodule is present in the thyroid ultrasound video, determine the video frame with the highest detection confidence of the thyroid nodule from video frames where the size of the thyroid nodule is larger than a set size, and mark the size of the thyroid nodule in the target video frame; when the thyroid nodule is not present in the thyroid ultrasound video, determine the video frame with the highest detection confidence of the thyroid gland from video frames that fully display the thyroid gland, and select it as the report frame.
[0013] A thyroid nodule detection system according to the present invention further includes: The report generation module is configured to generate a screening conclusion report corresponding to the thyroid ultrasound video based on the report frame and the detection results of the target thyroid nodule; The target thyroid nodule detection result is positive, suspected, negative, or fails quality control; a positive result indicates the presence of at least one definite nodule; a suspected result indicates the presence of a suspected nodule but no definite nodule; a negative result indicates the presence of neither a definite nor a suspected nodule; and a failed quality control result indicates that the quality of the thyroid ultrasound video does not meet diagnostic requirements.
[0014] This invention also provides a method for detecting thyroid nodules, comprising: Frame-by-frame detection is performed on target objects contained in thyroid ultrasound videos to obtain detection data for each target object in each video frame. The detection data includes segmentation mask, detection box, and detection confidence. Each target object includes carotid artery, thyroid gland, and thyroid nodule. Based on each of the detection boxes and each of the detection confidence levels, the tracking trajectories of each of the target objects are determined; Based on the tracking trajectories of the carotid artery and / or the tracking trajectories of the thyroid gland, the quality control of the thyroid ultrasound video is checked. If the quality control of the thyroid ultrasound video is qualified, a multi-level false positive filtering mechanism is adopted based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule to determine the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video. The multi-level false positive filtering mechanism includes an anatomical filtering mechanism, a continuity filtering mechanism and a morphological filtering mechanism.
[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the thyroid nodule detection method as described above.
[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the thyroid nodule detection method as described above.
[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the thyroid nodule detection method as described above.
[0018] The thyroid nodule detection system provided by this invention performs frame-by-frame detection of target objects in a thyroid ultrasound video, obtaining detection data for each target object in each video frame. The detection data includes a segmentation mask, a detection box, and a detection confidence level. Each target object includes a carotid artery, a thyroid gland, and a thyroid nodule. A trajectory determination module determines the tracking trajectory of each target object based on the detection boxes and the detection confidence levels. A video quality control module checks whether the thyroid ultrasound video meets quality control standards based on the tracking trajectories of the carotid artery and / or the thyroid gland. If the thyroid ultrasound video meets quality control standards, a nodule classification module, based on the segmentation mask of each target object and the tracking trajectories of the thyroid nodules, uses a multi-level false positive filtering mechanism to determine the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video. The multi-level false positive filtering mechanism includes an anatomical filtering mechanism, a continuity filtering mechanism, and a morphological filtering mechanism.
[0019] This invention reduces the false positive rate through a multi-level false positive filtering mechanism, thereby reducing unnecessary doctor reviews and patient anxiety, and enhancing stability and reliability. Diagnostic decisions based on multi-frame consistency are more reliable than single-frame diagnoses. Through quality control standardization, it automatically identifies unqualified videos, reduces ineffective work, ensures diagnostic consistency, and thus improves screening efficiency and diagnostic accuracy. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0021] Figure 1 This is one of the structural schematic diagrams of the thyroid nodule detection system provided by the present invention.
[0022] Figure 2 This is a schematic diagram of a report frame provided by the present invention.
[0023] Figure 3 This is the second schematic diagram of the thyroid nodule detection system provided by the present invention.
[0024] Figure 4 This is a flowchart illustrating the thyroid nodule detection method provided by the present invention.
[0025] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0027] First, a brief description of the relevant content involved in this invention will be given.
[0028] Some existing thyroid nodule identification systems can process complete ultrasound video sequences. Their technical approach involves first using a trained neural network model to extract nodule bounding boxes from multiple frames of the video. Then, by calculating the similarity (including positional similarity, size similarity, and temporal similarity) of the nodule bounding boxes between adjacent frames, they determine whether they belong to the same nodule, thus achieving the identification of multiple nodules and extraction of keyframes. While this system considers the temporal characteristics of the video, it lacks in-depth analysis of nodule stability and a false positive filtering mechanism.
[0029] Some existing thyroid nodule recognition systems propose the concept of dynamic input images, stitching together and resizing ultrasound images from three consecutive sampling points to form a dynamic input image. In terms of model architecture, sampling and feature extraction modules are added to the backbone network of the target detection model to reduce feature loss during network downsampling. This system also employs a complete intersection-union loss function, comprehensively considering factors such as overlap area, center point distance, and aspect ratio, making the predicted bounding box more accurate. However, this system primarily focuses on detection stability and lacks a complete diagnostic decision-making process.
[0030] Furthermore, most existing thyroid nodule identification systems assume good input data quality, but in actual screening, problems such as non-standard scanning, excessively fast probe movement, and poor image quality frequently occur. The systems lack the ability to automatically assess scan quality and cannot distinguish between real nodules and transient artifacts based on the stability of nodule appearance in the video.
[0031] Most existing thyroid nodule identification systems provide final diagnostic results in a "black box" manner, making it difficult for doctors to understand the system's decision-making basis. In screening scenarios, doctors need to know the specific reasons why the system judges a result as "positive" or "suspected," including quantitative indicators such as the size, location, and stability of the nodule.
[0032] To address at least one of the aforementioned problems, the present invention provides a thyroid nodule detection system.
[0033] The following is combined with Figures 1 to 5 The thyroid nodule detection system of the present invention is described.
[0034] Figure 1 This is one of the structural schematic diagrams of the thyroid nodule detection system provided by the present invention, such as... Figure 1 As shown, the thyroid nodule detection system includes the following: The frame-by-frame detection module 101 is configured to perform frame-by-frame detection of target objects contained in a thyroid ultrasound video, and obtain detection data of each target object in each video frame. The detection data includes a segmentation mask, a detection box, and a detection confidence score. Each target object includes a carotid artery, a thyroid gland, and a thyroid nodule. The trajectory determination module 102 is configured to determine each tracking trajectory of each target object based on each detection box and each detection confidence level; The video quality control module 103 is configured to detect whether the thyroid ultrasound video is qualified based on each tracking trajectory of the carotid artery and / or each tracking trajectory of the thyroid gland. The nodule classification module 104 is configured to, under the condition that the quality control of the thyroid ultrasound video is qualified, determine the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video by adopting a multi-level false positive filtering mechanism based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule. The multi-level false positive filtering mechanism includes an anatomical filtering mechanism, a continuity filtering mechanism and a morphological filtering mechanism.
[0035] Specifically, the anatomical filtering mechanism utilizes segmentation masks of the carotid artery and thyroid nodules to establish a safe distance model. This model determines the distance between the carotid artery and thyroid nodules, such as the center-to-center distance and the crossover ratio, to exclude detections with excessive overlap with the carotid artery. Furthermore, the anatomical filtering mechanism strictly restricts thyroid nodules to be located within the thyroid gland based on segmentation masks of the thyroid gland and thyroid nodules, and establishes prior knowledge of the thyroid's anatomical location to exclude areas where thyroid nodules are impossible.
[0036] Specifically, the target thyroid nodule detection result is positive, suspected, negative, or fails quality control; a positive result indicates the presence of at least one definite nodule; a suspected result indicates the presence of a suspected nodule but no definite nodule; a negative result indicates the presence of neither a definite nor a suspected nodule; and a failure to pass quality control indicates that the quality of the thyroid ultrasound video does not meet diagnostic requirements.
[0037] The continuous filtering mechanism takes advantage of the fact that real thyroid nodules must have a continuous and stable existence in space and time, and setting continuous frame detection can eliminate instantaneous noise.
[0038] A morphological filtering mechanism is used to set a reasonable range for thyroid nodule size, such as 3 mm to 35 mm, to filter thyroid nodules that are too large or too small.
[0039] In practical applications, users can input thyroid ultrasound videos that require thyroid nodule detection into the thyroid nodule detection system.
[0040] Then, the thyroid nodule detection system uses a frame-by-frame detection module to call a trained target instance segmentation model to perform frame-by-frame detection of the carotid artery, thyroid gland, and thyroid nodules in the thyroid ultrasound video, obtaining the detection results of the carotid artery, thyroid gland, and thyroid nodules in each video frame. The detection data includes detection boxes and confidence scores, and may also include pixel-level segmentation masks.
[0041] After obtaining the detection results of the carotid artery, thyroid gland and thyroid nodules in each video frame, the thyroid nodule detection system uses a trajectory determination module to track each target object based on the detection data of the target object in each video frame, and obtains at least one tracking trajectory for the target object.
[0042] Specifically, for each target object, the following steps are performed: For each video frame, predict the position of each existing trajectory of the target object in the video frame based on Kalman filtering; determine the detection box type of the target object in the video frame based on the confidence level of the target object; if the detection box type is a high-confidence detection box, associate the high-confidence detection box with each existing trajectory based on the position of each existing trajectory in the video frame; if the detection box type is a low-confidence detection box, associate the low-confidence detection box with each remaining unmatched trajectory based on the position of each existing unmatched trajectory in the video frame; if the detection box type is a high-confidence detection box and the association is unsuccessful, create a new trajectory for the target object based on the high-confidence detection box.
[0043] It should be noted that the trajectory determination module can also perform quality assessment on each of the tracking trajectories of each target object to obtain the target quality assessment result of each tracking trajectory of the target object; and filter the tracking trajectories whose target quality assessment results are not passed.
[0044] Furthermore, the thyroid nodule detection system uses a video quality control module to perform video quality control based on the tracking trajectories of the carotid artery and / or the thyroid gland, that is, to detect whether the thyroid ultrasound video is qualified or unqualified.
[0045] If the thyroid ultrasound video is found to be substandard, no further steps are required; the result of the target thyroid nodule test can be directly determined as substandard.
[0046] When the thyroid ultrasound video is deemed acceptable (i.e., the quality control of the thyroid ultrasound video is passed), the thyroid nodule detection system, through its nodule classification module, uses a multi-level false positive filtering mechanism based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule to determine the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video. This multi-level false positive filtering mechanism includes anatomical filtering, continuity filtering, and morphological filtering.
[0047] The thyroid nodule detection system provided by this invention can reduce the false positive rate through a multi-level false positive filtering mechanism, thereby reducing unnecessary doctor review and patient anxiety, enhancing stability and reliability. Diagnostic decisions based on multi-frame consistency are more reliable than single-frame diagnoses. Through quality control standardization, it automatically identifies unqualified videos, reduces ineffective work, ensures diagnostic consistency, and thus improves screening efficiency and diagnostic accuracy.
[0048] In one or more optional embodiments of the present invention, the video quality control module is specifically configured as follows: The tracking trajectories of the thyroid gland are arranged in descending order of length to obtain the first trajectory sequence; If a first target tracking trajectory exists in the first N tracking trajectories of the first trajectory sequence, the thyroid ultrasound video quality control is deemed qualified. If the first target tracking trajectory is not found in the first N tracking trajectories of the first trajectory sequence, the thyroid ultrasound video quality control is determined to be unqualified. Wherein, the first target tracking trajectory is a tracking trajectory that meets the first set quality control conditions; The first set quality control conditions include: the trajectory length of the tracking trajectory is greater than a length threshold, the average confidence value of the tracking trajectory is greater than a trajectory confidence threshold, and the number of video frames in which the detection box of the thyroid gland in the tracking trajectory is located in the center of the video frame is greater than a number threshold.
[0049] In practical applications, the lengths of all thyroid gland tracking trajectories obtained from the thyroid ultrasound video are first sorted. Only the longest N (e.g., 3) thyroid gland tracking trajectories are analyzed. At least one thyroid gland tracking trajectory must have a length greater than a length threshold (e.g., 30 frames), and the mean confidence score of that thyroid gland tracking trajectory must be greater than a trajectory confidence threshold (e.g., 0.6). Furthermore, the number of video frames in which the detection box of the thyroid gland is located in the center of the image within that thyroid gland tracking trajectory must be greater than a number threshold (e.g., 20 frames). In this case, the thyroid ultrasound video is determined to be a qualified video; otherwise, the thyroid ultrasound video is considered an unqualified video.
[0050] In this embodiment of the invention, by standardizing quality control, videos that fail to meet the standards are automatically identified, reducing unnecessary work; and a unified video quality assessment standard can ensure consistent diagnosis.
[0051] Optionally, the video quality control module is specifically configured as follows: The tracking trajectories of the carotid artery are arranged in descending order of length to obtain the second trajectory sequence; If a second target tracking trajectory exists in the first M tracking trajectories of the second trajectory sequence, it is determined that the thyroid ultrasound video has passed the quality control test; If the second target tracking trajectory is not present in the first M tracking trajectories of the second trajectory sequence, it is determined that the thyroid ultrasound video has failed the quality control test; The second target tracking trajectory is a tracking trajectory that meets the second set quality control conditions; The second set quality control conditions include: the trajectory length of the tracking trajectory is greater than a length threshold, the average detection confidence value of the tracking trajectory is greater than a trajectory detection confidence threshold, and the number of video frames in which the detection box of the carotid artery in the tracking trajectory is located outside the thyroid gland in the video frame is greater than a number threshold.
[0052] Optionally, the video quality control module is specifically configured as follows: The tracking trajectories of the thyroid gland are arranged in descending order to obtain the first trajectory sequence, and the tracking trajectories of the carotid artery are arranged in descending order to obtain the second trajectory sequence. If a first target tracking trajectory exists in the first N tracking trajectories of the first trajectory sequence, and a second target tracking trajectory exists in the first M tracking trajectories of the second trajectory sequence, then the thyroid ultrasound video quality control is deemed qualified. If the first target tracking trajectory is not present in the first N tracking trajectories of the first trajectory sequence, and / or the second target tracking trajectory is not present in the first M tracking trajectories of the second trajectory sequence, then the thyroid ultrasound video is determined to have failed the quality control test.
[0053] Optionally, the nodule classification module is specifically configured as follows: Based on the aforementioned anatomical filtering mechanism, the aforementioned continuity filtering mechanism, and the aforementioned morphological filtering mechanism, criteria for identifying nodules, suspected nodules, and false positive nodules are established. Based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule, the target parameters of the thyroid nodule are determined; The target parameters are compared with the criteria for confirmed nodules, the criteria for suspected nodules, and the criteria for false positive nodules, respectively, to obtain the detection results of the target thyroid nodules corresponding to the thyroid ultrasound video.
[0054] Specifically, the anatomical filtering mechanism indicates that the thyroid nodule and the carotid artery do not overlap and that the thyroid nodule is located inside the thyroid gland; the continuity filtering mechanism indicates that the number of consecutive frames of the thyroid nodule is greater than or equal to a first frame number threshold and the mean of the detection confidence of the thyroid nodule is greater than or equal to a first mean threshold; the morphological filtering mechanism indicates that the average size of the thyroid nodule is greater than or equal to a size threshold.
[0055] Specifically, the target parameters include the number of consecutive frames of thyroid nodules, the mean confidence score of thyroid nodules, the average size of thyroid nodules, the positional relationship between thyroid nodules and thyroid glands, and whether there is an overlapping area between thyroid nodules and carotid arteries.
[0056] Specifically, the criteria for determining a nodule refer to the standards or conditions characterizing the existence of a definite nodule. These criteria include: the number of consecutive frames of the thyroid nodule's occurrence is greater than or equal to a first frame count threshold; the mean detection confidence score of the thyroid nodule is greater than or equal to a first mean threshold; the average size of the thyroid nodule is greater than or equal to a size threshold; the thyroid nodule is located within the thyroid gland; and the thyroid nodule does not overlap with the carotid artery.
[0057] Among them, the number of consecutive nodule appearance frames refers to the number of video frames in the thyroid ultrasound video in which thyroid nodules are detected consecutively; the average size of the thyroid nodule can be the average area of the thyroid nodule, the average of the transverse or longitudinal diameter of the thyroid nodule, or the average of the maximum values of the transverse and longitudinal diameters of the thyroid nodule.
[0058] For example, the criteria for identifying a nodule are: the number of consecutive occurrences ≥3 frames, the average detection confidence level ≥0.7, the average of the maximum values of the transverse and longitudinal diameters ≥3mm, the location within the tissue of the thyroid gland, and the absence of overlap with the carotid artery.
[0059] Specifically, the suspected nodule criteria refer to the standards used to characterize the presence of a suspected nodule. The suspected nodule criteria meet most of the conditions in the nodule determination criteria, but some indicators are slightly below a threshold. The suspected nodule criteria include: the thyroid nodule is located within the thyroid gland; the thyroid nodule does not overlap with the carotid artery; at least one of the following parameters—the number of consecutive nodule occurrences, the mean detection confidence score of the thyroid nodule, and the average size of the thyroid nodule—is less than the threshold corresponding to that parameter, and the difference between the threshold and the parameter is less than a set value corresponding to the parameter.
[0060] Among them, the set values are slightly smaller than: for example, the set value for the number of consecutive frames of nodules is 1 frame, and the set value for the mean of detection confidence is 0.5.
[0061] Specifically, the false positive nodule criteria refer to the standards or conditions characterizing the presence of false positive nodules. The false positive nodule criteria include at least one of the following: the overlap between the thyroid nodule and the carotid artery is greater than an overlap threshold; the thyroid nodule is located outside the thyroid gland; the number of consecutive frames of the thyroid nodule is less than a second frame number threshold; and / or, the mean detection confidence score of the thyroid nodule is less than a second mean threshold.
[0062] For example, the overlap between the thyroid nodule and the carotid artery region is greater than the overlap threshold, and / or the thyroid nodule is located entirely outside the thyroid gland tissue, and / or the number of frames in which the thyroid nodule is present is too low or the detection confidence is too low.
[0063] In practical applications, criteria for identifying nodules, suspected nodules, and false positive nodules can be established based on anatomical, continuity, and morphological filtering mechanisms. Then, based on the segmentation mask of each target object and the tracking trajectories of each thyroid nodule, target parameters for the thyroid nodule are determined, including the number of consecutive nodule occurrences, the mean detection confidence score of the thyroid nodule, the average size of the thyroid nodule, the positional relationship between the thyroid nodule and the thyroid gland, and whether there is an overlapping area between the thyroid nodule and the carotid artery.
[0064] Furthermore, the target parameters were compared with the criteria for confirmed nodules, suspected nodules, and false positive nodules: if the target parameters met the criteria for confirmed nodules, the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video was positive, indicating the presence of at least one confirmed nodule; if the target parameters met the criteria for suspected nodules, the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video was suspected, indicating the presence of suspected nodules but no confirmed nodules; if the target parameters met the criteria for false positive nodules, the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video was negative, indicating the presence of neither confirmed nor suspected nodules.
[0065] In this embodiment of the invention, by establishing criteria for identifying nodules, suspected nodules, and false positive nodules, the false positive rate is significantly reduced. Through multi-level filtering, the false positive rate is reduced by more than 50% compared to the single-frame detection system, reducing unnecessary doctor review and patient anxiety. Stability and reliability are enhanced, and diagnostic decisions based on multi-frame consistency are more reliable than single-frame diagnosis.
[0066] Optionally, the system further includes: The nodule measurement and statistics module is configured to determine the minimum bounding rectangle of each segmentation mask of the thyroid nodule; determine the size of the thyroid nodule based on the width and height of each minimum bounding rectangle; count the number of thyroid nodules in the thyroid ultrasound video, and classify the thyroid nodules according to their size.
[0067] In practical applications, the thyroid nodule detection system includes a structured screening report generation system, which contains a nodule measurement and statistics module.
[0068] The nodule measurement and statistics module takes the segmentation mask of the thyroid nodule as input, calculates the width and height of the minimum bounding rectangle of the segmentation mask, and the corresponding measurement values of the size of the thyroid nodule, i.e. the measurement values of the horizontal and vertical axes; it counts the total number of thyroid nodules appearing in the video and classifies them by size.
[0069] In this embodiment of the invention, the accuracy of screening can be improved by using a nodule measurement and statistics module.
[0070] Optionally, the system further includes: The report frame selection module is configured to, when the thyroid nodule is present in the thyroid ultrasound video, determine the video frame with the highest detection confidence of the thyroid nodule from video frames where the size of the thyroid nodule is larger than a set size, and mark the size of the thyroid nodule in the target video frame; when the thyroid nodule is not present in the thyroid ultrasound video, determine the video frame with the highest detection confidence of the thyroid gland from video frames that fully display the thyroid gland, and select it as the report frame.
[0071] Specifically, a report frame refers to a video frame used for presentation in a report.
[0072] In practical applications, the structured screening report generation system also includes a report frame selection module, which calls a report frame selection algorithm to determine the report frame.
[0073] The report frame selection algorithm is as follows: For videos with nodules (thyroid ultrasound videos with thyroid nodules): for each identified nodule (the target thyroid nodule detection result of the thyroid ultrasound video is positive), the video frame with the highest detection confidence among the larger video frames in its trajectory is selected as the report frame; for normal videos (thyroid ultrasound videos without thyroid nodules): the video frame with the highest confidence among the video frames showing a relatively complete thyroid is selected as the report frame.
[0074] See Figure 2 , Figure 2 This is a schematic diagram of a report frame provided by the present invention: where “Width” represents the horizontal axis length (horizontal diameter) of the thyroid nodule, and “Height” represents the vertical axis length (vertical diameter) of the thyroid nodule.
[0075] In this embodiment of the invention, the report frame selection algorithm can select the most prominent video frame in the most prominent thyroid nodule segment as the report frame, so as to facilitate review by medical staff, reduce the workload of review, and improve screening efficiency.
[0076] Optionally, the system further includes: The report generation module is configured to generate a screening conclusion report corresponding to the thyroid ultrasound video based on the report frame and the detection results of the target thyroid nodule.
[0077] The structured screening report generation system also includes a report generation module, which can summarize the report frame and the detection results of the target thyroid nodules to generate a screening conclusion report for medical staff to view.
[0078] The following is combined with Figure 3 The thyroid nodule detection system provided by the present invention will be further described.
[0079] Figure 3 This is a second schematic diagram of the thyroid nodule detection system provided by the present invention, as shown below. Figure 3 As shown, the processing flow of this thyroid nodule detection system is as follows.
[0080] First, the user inputs a thyroid ultrasound video into the thyroid nodule detection system. Then, the thyroid nodule detection system performs frame-by-frame detection on the thyroid ultrasound video using a frame-by-frame detection module, obtaining the segmentation mask, detection box, and detection confidence score for any target object among the carotid artery, thyroid gland, and thyroid nodules.
[0081] Then, the trajectory determination module performs Kalman filtering prediction and two-stage association matching to obtain the tracking trajectory of each target object. Furthermore, the trajectory determination module performs quality evaluation on each tracking trajectory, including indicators such as trajectory length, continuity, and motion smoothness. A trajectory confidence scoring mechanism is established to filter out transient and unstable detections.
[0082] Next, intelligent post-processing is performed. The video quality control module determines whether the thyroid ultrasound video passes quality control: the lengths of all thyroid gland tracking trajectories obtained in the thyroid ultrasound video are sorted, and only the first three thyroid trajectories are analyzed. If the length of at least one thyroid trajectory is greater than a threshold (e.g., 30 frames), and the average confidence of that trajectory is greater than a threshold (e.g., 0.6), and the number of video frames in which the thyroid detection box is located in the center of the image in the trajectory is greater than a certain threshold (e.g., 20 frames), then the quality control is passed. Otherwise, the thyroid ultrasound video fails quality control and needs to be output as "Quality Control Failed," meaning the target thyroid nodule detection result is unqualified.
[0083] Furthermore, if the quality control is passed, the nodule classification module uses the nodule classification engine to perform multi-level false positive filtering to obtain the detection results of the target thyroid nodules.
[0084] Finally, a structured report is generated. Based on the target thyroid nodule detection results and the nodule classification engine, nodule measurement and statistics are performed, and report frames are selected to generate a structured screening report; alternatively, if quality control fails, a structured screening report is generated based on the target thyroid nodule detection results. The structured screening report is then output.
[0085] The thyroid nodule detection system provided by this invention can significantly improve screening efficiency: the degree of automation is greatly improved, realizing end-to-end automated processing from raw video to structured report; it supports batch processing, making it suitable for large-scale screening scenarios; quality control is standardized, automatically identifying unqualified videos and reducing unnecessary work; and a unified image quality assessment standard ensures diagnostic consistency.
[0086] The thyroid nodule detection system provided by this invention can significantly improve diagnostic accuracy: the false positive rate is significantly reduced, and through multi-level filtering, the false positive rate is reduced by more than 50% compared with the single-frame detection system, reducing unnecessary doctor review and patient anxiety; the stability and reliability are enhanced, and the diagnostic decision based on multi-frame consistency is more reliable than single-frame diagnosis.
[0087] The thyroid nodule detection system provided by this invention can comprehensively improve clinical usability: Report standardization and structuring automatically generate structured reports that conform to clinical standards, including all necessary diagnostic elements: quantity, size, and location; supports the Digital Imaging and Communications in Medicine (DICOM) standard interface, facilitating integration with hospital information systems; the decision-making process is interpretable, providing detailed diagnostic evidence and quantitative indicators, allowing doctors to trace the system's judgment logic, enhancing trust, and facilitating quality control and continuous improvement; adaptable to different screening scenarios, with adjustable parameters to meet different needs from community screening to specialist diagnosis; supports multiple ultrasound devices and scanning protocols, possessing good generalization ability and robustness.
[0088] The thyroid nodule detection method provided by the present invention is described below. The thyroid nodule detection method described below can be referred to in correspondence with the thyroid nodule detection system described above.
[0089] Figure 4 This is a schematic flowchart of the thyroid nodule detection method provided by the present invention, as shown below. Figure 4 As shown, the method for detecting thyroid nodules includes the following: Step 401: Detect the target objects in the thyroid ultrasound video frame by frame to obtain the detection data of each target object in each video frame. The detection data includes a segmentation mask, a detection box, and a detection confidence score. Each target object includes the carotid artery, thyroid gland, and thyroid nodule. Step 402: Based on each detection box and each detection confidence level, determine each tracking trajectory of each target object; Step 403: Based on the tracking trajectories of the carotid artery and / or the tracking trajectories of the thyroid gland, determine whether the thyroid ultrasound video meets quality control standards; Step 404: If the quality control of the thyroid ultrasound video is qualified, based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule, a multi-level false positive filtering mechanism is adopted to determine the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video. The multi-level false positive filtering mechanism includes an anatomical filtering mechanism, a continuity filtering mechanism and a morphological filtering mechanism.
[0090] The thyroid nodule detection method provided by this invention can reduce the false positive rate through a multi-level false positive filtering mechanism, reduce unnecessary doctor review and patient anxiety, and enhance stability and reliability. Diagnostic decisions based on multi-frame consistency are more reliable than single-frame diagnoses. Through quality control standardization, it automatically identifies unqualified videos, reduces invalid work, ensures diagnostic consistency, and thus improves screening efficiency and diagnostic accuracy.
[0091] Optionally, the step of detecting whether the thyroid ultrasound video meets quality control standards based on each tracking trajectory of the carotid artery and / or each tracking trajectory of the thyroid gland includes: The tracking trajectories of the thyroid gland are arranged in descending order of length to obtain the first trajectory sequence; If a first target tracking trajectory exists in the first N tracking trajectories of the first trajectory sequence, the thyroid ultrasound video quality control is deemed qualified. If the first target tracking trajectory is not found in the first N tracking trajectories of the first trajectory sequence, the thyroid ultrasound video quality control is determined to be unqualified. Wherein, the first target tracking trajectory is a tracking trajectory that meets the first set quality control conditions; The first set quality control conditions include: the trajectory length of the tracking trajectory is greater than a length threshold, the average confidence value of the tracking trajectory is greater than a trajectory confidence threshold, and the number of video frames in which the detection box of the thyroid gland in the tracking trajectory is located in the center of the video frame is greater than a number threshold.
[0092] Optionally, based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule, a multi-level false positive filtering mechanism is used to determine the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video, including: Based on the aforementioned anatomical filtering mechanism, the aforementioned continuity filtering mechanism, and the aforementioned morphological filtering mechanism, criteria for identifying nodules, suspected nodules, and false positive nodules are established. Based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule, the target parameters of the thyroid nodule are determined; The target parameters are compared with the criteria for confirmed nodules, the criteria for suspected nodules, and the criteria for false positive nodules, respectively, to obtain the detection results of the target thyroid nodules corresponding to the thyroid ultrasound video.
[0093] Optionally, the criteria for determining nodules include: The number of consecutive frames in which the thyroid nodules appear is greater than or equal to the first frame number threshold; The mean confidence level of the detection of the thyroid nodules is greater than or equal to a first mean threshold. The average size of the thyroid nodules is greater than or equal to the size threshold; The thyroid nodule is located inside the thyroid gland; Furthermore, the thyroid nodule and the carotid artery do not overlap.
[0094] Optionally, the criteria for suspected nodules include: The thyroid nodule is located inside the thyroid gland; The thyroid nodule and the carotid artery do not overlap. At least one of the following parameters—the number of consecutive frames of nodules in the thyroid nodule, the mean of the detection confidence of the thyroid nodule, and the average size of the thyroid nodule—is less than the threshold corresponding to the parameter, and the difference between the threshold corresponding to the parameter and the parameter is less than the set value corresponding to the parameter.
[0095] Optionally, the criteria for false positive nodules include at least one of the following: The overlap between the thyroid nodule and the carotid artery is greater than the overlap threshold; The thyroid nodule is located outside the thyroid gland; The number of consecutive frames of the thyroid nodules is less than a second frame number threshold, and / or the mean of the detection confidence of the thyroid nodules is less than a second mean threshold.
[0096] Optionally, the anatomical filtering mechanism characterizes that the thyroid nodule and the carotid artery do not overlap and that the thyroid nodule is located inside the thyroid gland; The continuous filtering mechanism is characterized by the number of consecutive frames of the thyroid nodules being greater than or equal to a first frame number threshold and the mean of the detection confidence of the thyroid nodules being greater than or equal to a first mean threshold. The morphological filtering mechanism characterizes the average size of the thyroid nodules as being greater than or equal to a size threshold.
[0097] Optionally, the method further includes: The nodule measurement and statistics module is configured to determine the minimum bounding rectangle of each segmentation mask of the thyroid nodule; determine the size of the thyroid nodule based on the width and height of each minimum bounding rectangle; count the number of thyroid nodules in the thyroid ultrasound video, and classify the thyroid nodules according to their size.
[0098] Optionally, the method further includes: The report frame selection module is configured to, when the thyroid nodule is present in the thyroid ultrasound video, determine the video frame with the highest detection confidence of the thyroid nodule from video frames where the size of the thyroid nodule is larger than a set size, and mark the size of the thyroid nodule in the target video frame; when the thyroid nodule is not present in the thyroid ultrasound video, determine the video frame with the highest detection confidence of the thyroid gland from video frames that fully display the thyroid gland, and select it as the report frame.
[0099] Optionally, the method further includes: The report generation module is configured to generate a screening conclusion report corresponding to the thyroid ultrasound video based on the report frame and the detection results of the target thyroid nodule; The target thyroid nodule detection result is positive, suspected, negative, or fails quality control; a positive result indicates the presence of at least one definite nodule; a suspected result indicates the presence of a suspected nodule but no definite nodule; a negative result indicates the presence of neither a definite nor a suspected nodule; and a failed quality control result indicates that the quality of the thyroid ultrasound video does not meet diagnostic requirements.
[0100] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 5 As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a thyroid nodule detection method.
[0101] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0102] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the thyroid nodule detection methods provided by the above methods.
[0103] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the thyroid nodule detection methods provided by the methods described above.
[0104] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0105] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A thyroid nodule detection system, characterized in that, include: The frame-by-frame detection module is configured to perform frame-by-frame detection of target objects contained in a thyroid ultrasound video, and obtain detection data of each target object in each video frame. The detection data includes a segmentation mask, a detection box, and a detection confidence score. Each target object includes a carotid artery, a thyroid gland, and a thyroid nodule. The trajectory determination module is configured to determine each tracking trajectory of each target object based on each detection box and each detection confidence level; The video quality control module is configured to detect whether the thyroid ultrasound video is qualified based on each tracking trajectory of the carotid artery and / or each tracking trajectory of the thyroid gland; The nodule classification module is configured to, under the condition that the quality control of the thyroid ultrasound video is qualified, determine the detection result of the target thyroid nodule corresponding to the thyroid ultrasound video by adopting a multi-level false positive filtering mechanism based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule. The multi-level false positive filtering mechanism includes an anatomical filtering mechanism, a continuity filtering mechanism, and a morphological filtering mechanism.
2. The thyroid nodule detection system according to claim 1, characterized in that, The video quality control module is specifically configured as follows: The tracking trajectories of the thyroid gland are arranged in descending order of length to obtain the first trajectory sequence; If a first target tracking trajectory exists in the first N tracking trajectories of the first trajectory sequence, the thyroid ultrasound video quality control is deemed qualified. If the first target tracking trajectory is not found in the first N tracking trajectories of the first trajectory sequence, the thyroid ultrasound video quality control is determined to be unqualified. Wherein, the first target tracking trajectory is a tracking trajectory that meets the first set quality control conditions; The first set quality control conditions include: the trajectory length of the tracking trajectory is greater than a length threshold, the average confidence value of the tracking trajectory is greater than a trajectory confidence threshold, and the number of video frames in which the detection box of the thyroid gland in the tracking trajectory is located in the center of the video frame is greater than a number threshold.
3. The thyroid nodule detection system according to claim 1, characterized in that, The nodule classification module is specifically configured as follows: Based on the aforementioned anatomical filtering mechanism, the aforementioned continuity filtering mechanism, and the aforementioned morphological filtering mechanism, criteria for identifying nodules, suspected nodules, and false positive nodules are established. Based on the segmentation mask of each target object and the tracking trajectory of each thyroid nodule, the target parameters of the thyroid nodule are determined; The target parameters are compared with the criteria for confirmed nodules, the criteria for suspected nodules, and the criteria for false positive nodules, respectively, to obtain the detection results of the target thyroid nodules corresponding to the thyroid ultrasound video.
4. The thyroid nodule detection system according to claim 3, characterized in that, The criteria for determining nodules include: The number of consecutive frames in which the thyroid nodules appear is greater than or equal to the first frame number threshold; The mean confidence level of the detection of the thyroid nodules is greater than or equal to a first mean threshold. The average size of the thyroid nodules is greater than or equal to the size threshold; The thyroid nodule is located inside the thyroid gland; Furthermore, the thyroid nodule and the carotid artery do not overlap.
5. The thyroid nodule detection system according to claim 4, characterized in that, The criteria for suspected nodules include: The thyroid nodule is located inside the thyroid gland; The thyroid nodule and the carotid artery do not overlap. At least one of the following parameters—the number of consecutive frames of nodules in the thyroid nodule, the mean of the detection confidence of the thyroid nodule, and the average size of the thyroid nodule—is less than the threshold corresponding to the parameter, and the difference between the threshold corresponding to the parameter and the parameter is less than the set value corresponding to the parameter.
6. The thyroid nodule detection system according to claim 3, characterized in that, The criteria for false positive nodules include at least one of the following: The overlap between the thyroid nodule and the carotid artery is greater than the overlap threshold; The thyroid nodule is located outside the thyroid gland; The number of consecutive frames of the thyroid nodules is less than a second frame number threshold, and / or the mean of the detection confidence of the thyroid nodules is less than a second mean threshold.
7. The thyroid nodule detection system according to claim 3, characterized in that, The anatomical filtering mechanism indicates that the thyroid nodule and the carotid artery do not overlap and that the thyroid nodule is located inside the thyroid gland. The continuous filtering mechanism is characterized by the number of consecutive frames of the thyroid nodules being greater than or equal to a first frame number threshold and the mean of the detection confidence of the thyroid nodules being greater than or equal to a first mean threshold. The morphological filtering mechanism characterizes the average size of the thyroid nodules as being greater than or equal to a size threshold.
8. The thyroid nodule detection system according to any one of claims 1-7, characterized in that, The system also includes: The nodule measurement and statistics module is configured to determine the minimum bounding rectangle of each segmentation mask of the thyroid nodule; determine the size of the thyroid nodule based on the width and height of each minimum bounding rectangle; count the number of thyroid nodules in the thyroid ultrasound video, and classify the thyroid nodules according to their size.
9. The thyroid nodule detection system according to claim 8, characterized in that, The system also includes: The report frame selection module is configured to, when the thyroid nodule is present in the thyroid ultrasound video, determine the video frame with the highest detection confidence of the thyroid nodule from video frames where the size of the thyroid nodule is larger than a set size, and mark the size of the thyroid nodule in the target video frame; when the thyroid nodule is not present in the thyroid ultrasound video, determine the video frame with the highest detection confidence of the thyroid gland from video frames that fully display the thyroid gland, and select it as the report frame.
10. The thyroid nodule detection system according to claim 9, characterized in that, The system also includes: The report generation module is configured to generate a screening conclusion report corresponding to the thyroid ultrasound video based on the report frame and the detection results of the target thyroid nodule; The target thyroid nodule detection result is positive, suspected, negative, or fails quality control; a positive result indicates the presence of at least one definite nodule; a suspected result indicates the presence of a suspected nodule but no definite nodule; a negative result indicates the presence of neither a definite nor a suspected nodule; and a failed quality control result indicates that the quality of the thyroid ultrasound video does not meet diagnostic requirements.