CN-122023259-A - Thyroid nodule detection system

CN122023259ACN 122023259 ACN122023259 ACN 122023259ACN-122023259-A

Abstract

The invention provides a thyroid nodule detection system which comprises a receiving module, a multi-target instance segmentation module and a determining module, wherein the receiving module is configured to receive a thyroid ultrasonic video, the multi-target instance segmentation module is configured to detect target objects in video frames frame by frame through a trained target instance segmentation model to obtain detection data of the target objects in the video frames, the detection data comprise detection frames and confidence levels, the target objects comprise carotid arteries, thyroid glands and thyroid nodules, the instance tracking module is configured to track the target objects in the thyroid ultrasonic video by adopting a multi-target tracking algorithm based on target detection based on the detection data to obtain tracking tracks of the target objects, and the determining module is configured to determine target thyroid nodule detection results corresponding to the thyroid ultrasonic video based on the tracking tracks of the target objects. The invention can process the video aiming at the end-to-end screening scene and improve the detection accuracy.

Inventors

HAO MINGRUI
WANG SHUANGYI
GU XIAOLIN

Assignees

聆数医疗科技(苏州)有限公司
中国科学院自动化研究所

Dates

Publication Date: 20260512
Application Date: 20251229

Claims (10)

1. A thyroid nodule detection system, comprising: a receiving module configured to receive a thyroid ultrasound video, the thyroid ultrasound video comprising a plurality of video frames; The multi-target instance segmentation module is configured to detect target objects in each video frame by frame through a trained target instance segmentation model to obtain detection data of each target object in each video frame, wherein the detection data comprises a detection frame and confidence, and each target object comprises carotid artery, thyroid gland and thyroid nodule; the example tracking module is configured to track each target object in the thyroid ultrasonic video by adopting a multi-target tracking algorithm based on target detection based on each detection data to obtain each tracking track of each target object; And the determining module is configured to determine a target thyroid nodule detection result corresponding to the thyroid ultrasonic video based on each tracking track of each target object.
2. The thyroid nodule detection system of claim 1, further comprising a trajectory management module configured to: performing quality evaluation on each tracking track of the target object aiming at each target object to obtain a target quality evaluation result of each tracking track of the target object; and filtering the tracking track which is not passed by the target quality evaluation result.
3. The thyroid nodule detection system of claim 2, wherein the quality assessment comprises a first quality assessment; The track management module is specifically configured to: for each target object, performing the first quality evaluation on each tracking track of the target object to obtain a first quality evaluation result of each tracking track of the target object, wherein the index of the first quality evaluation comprises at least one of track length, continuity and motion smoothness; the target quality assessment results for each of the tracked trajectories of the target object are determined based on the first quality assessment results for each of the tracked trajectories of the target object.
4. The thyroid nodule detection system of claim 3, wherein the trajectory management module is specifically configured to: For each tracking track of the target object, acquiring the number of video frames corresponding to the tracking track and representing the track length, and determining a track length evaluation result of the tracking track based on the number of video frames; Acquiring an interval frame number corresponding to the tracking track and representing continuity, and determining a continuity evaluation result of the tracking track based on the interval frame number; Acquiring a continuous change rate corresponding to the tracking track and representing motion smoothness, and determining a motion smoothness evaluation result of the tracking track based on the continuous change rate, wherein the continuous change rate is the change rate of the overlapping degree of the detection frame and/or the change rate of the position of the detection frame in continuous video frames; The first quality assessment result of the tracking track is determined based on at least one of the track length assessment result, the continuity assessment result, and the motion smoothness assessment result.
5. The thyroid nodule detection system of claim 2, wherein the quality assessment comprises a second quality assessment; The track management module is specifically configured to: for each target object, acquiring a confidence coefficient sequence corresponding to each tracking track of the target object, wherein the confidence coefficient sequence comprises confidence coefficients corresponding to all detection frames corresponding to the tracking tracks; based on the confidence sequences corresponding to the tracking tracks of the target object, performing the second quality evaluation on the tracking tracks of the target object to obtain a second quality evaluation result of the tracking tracks of the target object, wherein the index of the second quality evaluation comprises at least one of a confidence mean value, a confidence minimum value and a confidence volatility; the target quality assessment results for each of the tracked trajectories of the target object are determined based on the second quality assessment results for each of the tracked trajectories of the target object.
6. The thyroid nodule detection system of claim 3, wherein the trajectory management module is specifically configured to: determining a confidence score for each of the tracking trajectories of the target object based on the first quality assessment result and/or the second quality assessment result; Determining that the target quality assessment result of the tracking track with the confidence score smaller than a scoring threshold value is failed, and determining that the target quality assessment result of the tracking track with the confidence score greater than or equal to the scoring threshold value is passed.
7. The thyroid nodule detection system of claim 1, wherein the instance tracking module is specifically configured to: for each of the target objects, performing the steps of: predicting, for each of the video frames, a position of each existing trajectory of the target object in the video frame based on kalman filtering; determining a detection frame type of the target object in the video frame based on the confidence level of the target object in the video frame; If the detection frame type is a high-confidence detection frame, associating the high-confidence detection frame with each existing track based on the position of each existing track in the video frame; under the condition that the detection frame type is a low-confidence detection frame, associating the low-confidence detection frame with each remaining unmatched track based on the position of the remaining unmatched track in each existing track in the video frame; and under the condition that the detection frame type is the high-confidence detection frame and the association is not successful, creating a new track for the target object based on the high-confidence detection frame.
8. The thyroid nodule detection system of claim 1, further comprising a training module configured to: Acquiring a plurality of initial images containing a specified object, wherein the specified object is at least one of carotid artery, thyroid gland and thyroid nodule; Labeling each appointed object in each initial image, and carrying out data enhancement on each initial image to obtain a training image set; and training the initial instance segmentation model based on the training image set to obtain the target instance segmentation model.
9. The thyroid nodule detection system of claim 8, wherein the training module is specifically configured to: And carrying out random brightness contrast adjustment and/or Gaussian noise addition on each initial image to realize data enhancement and obtain the training image set, wherein the Gaussian noise addition comprises simulation ultrasonic artifacts.
10. The thyroid nodule detection system of any one of claims 1-9, wherein the detection data further comprises a pixel level segmentation mask; the determining module is specifically configured to: Determining an initial thyroid nodule detection result based on each of the tracked trajectories of the thyroid nodules; and performing false positive filtering processing on the initial thyroid nodule detection result based on at least one of the tracking tracks of the carotid artery, the pixel-level segmentation masks of the carotid artery, the tracking tracks of the thyroid gland, the pixel-level segmentation masks of the thyroid gland and the pixel-level segmentation masks of the thyroid nodule to obtain the target thyroid nodule detection result.

Description

Thyroid nodule detection system Technical Field The invention relates to the technical field of image processing and medical equipment, in particular to a thyroid nodule detection system. Background The current thyroid nodule diagnosis system is mainly a classification diagnosis system based on a single frame image and a report generation system relying on front segmentation. A single frame image based classification diagnostic system is essentially based on analysis of static images and cannot process timing information in video sequences, whereas a pre-segmentation dependent report generation system workflow relies heavily on an independent and pre-nodule detection and segmentation module, and the overall system is not a true "end-to-end" screening solution. Therefore, an effective solution is needed to solve the above-mentioned problems. Disclosure of Invention In order to solve the problems, the invention provides a thyroid nodule detection system. The invention provides a thyroid nodule detection system, comprising: a receiving module configured to receive a thyroid ultrasound video, the thyroid ultrasound video comprising a plurality of video frames; The multi-target instance segmentation module is configured to detect target objects in each video frame by frame through a trained target instance segmentation model to obtain detection data of each target object in each video frame, wherein the detection data comprises a detection frame and confidence, and each target object comprises carotid artery, thyroid gland and thyroid nodule; the example tracking module is configured to track each target object in the thyroid ultrasonic video by adopting a multi-target tracking algorithm based on target detection based on each detection data to obtain each tracking track of each target object; a determining module configured to determine a target thyroid nodule detection result corresponding to the thyroid ultrasound video based on each tracking track of each target object The thyroid nodule detection system provided by the invention further comprises a track management module configured to: performing quality evaluation on each tracking track of the target object aiming at each target object to obtain a target quality evaluation result of each tracking track of the target object; and filtering the tracking track which is not passed by the target quality evaluation result. According to the thyroid nodule detection system provided by the invention, the quality assessment comprises a first quality assessment; The track management module is specifically configured to: for each target object, performing the first quality evaluation on each tracking track of the target object to obtain a first quality evaluation result of each tracking track of the target object, wherein the index of the first quality evaluation comprises at least one of track length, continuity and motion smoothness; the target quality assessment results for each of the tracked trajectories of the target object are determined based on the first quality assessment results for each of the tracked trajectories of the target object. According to the thyroid nodule detection system provided by the invention, the track management module is specifically configured to: For each tracking track of the target object, acquiring the number of video frames corresponding to the tracking track and representing the track length, and determining a track length evaluation result of the tracking track based on the number of video frames; Acquiring an interval frame number corresponding to the tracking track and representing continuity, and determining a continuity evaluation result of the tracking track based on the interval frame number; Acquiring a continuous change rate corresponding to the tracking track and representing motion smoothness, and determining a motion smoothness evaluation result of the tracking track based on the continuous change rate, wherein the continuous change rate is the change rate of the overlapping degree of the detection frame and/or the change rate of the position of the detection frame in continuous video frames; The first quality assessment result of the tracking track is determined based on at least one of the track length assessment result, the continuity assessment result, and the motion smoothness assessment result. According to the thyroid nodule detection system provided by the invention, the quality assessment comprises a second quality assessment; The track management module is specifically configured to: for each target object, acquiring a confidence coefficient sequence corresponding to each tracking track of the target object, wherein the confidence coefficient sequence comprises confidence coefficients corresponding to all detection frames corresponding to the tracking tracks; based on the confidence sequences corresponding to the tracking tracks of the target object, performing the second quality evaluation on the tracking t