Ultrasound image detection system and method thereof based on artificial intelligence (AI) automatic labeling of anatomical structures

1 . An ultrasound image detection system based on artificial intelligence automatic labeling of an anatomical structure, the ultrasound image detection system comprising:

a receiving module, executed by a processor, configured to receive an image to be recognized from an ultrasound apparatus;

an image recognition module, executed by the processor, having an object detection model and communicatively connected to the receiving module to receive the image to be recognized, such that the object detection model is used to detect a plurality of anatomical structures in the image to be recognized, thereby obtaining a plurality of object recognition images with object detection results;

an image processing module, executed by the processor, communicatively connected to the image recognition module to receive the plurality of object recognition images, wherein the image processing module uses at least one of the plurality of object recognition images received first as at least one reference image and the other one of the plurality of object recognition images received subsequently as an image to be recognized, such that a new confidence index of the plurality of anatomical structures in the image to be recognized is computed according to the object detection results of the image to be recognized and the reference image, a decay coefficient and a distance between the reference image and the image to be recognized so as to label missed anatomical structures in the image to be recognized, thereby detecting the missed anatomical structures and outputting an object detection image; and

a display module, executed by the processor, communicatively connected to the image processing module to receive the object detection image and display the object detection image in real time.

2 . The ultrasound image detection system of claim 1 , further comprising an automatic labeling module, executed by the processor, communicatively connected to the image recognition module and configured to use a semi-supervised learning to train a teacher model and a student model, such that the automatic labeling module uses the trained teacher model to label unlabeled plural ultrasound images to generate a plurality of training images, thereby providing for the image recognition module to train the object detection model.

3 . The ultrasound image detection system of claim 1 , wherein the object detection model includes a Focus framework which performs a slicing operation on the received image to be recognized, such that a feature map of the image to be recognized is obtained, thereby providing for the object detection model to detect a plurality of anatomical structures in the image to be recognized.

4 . The ultrasound image detection system of claim 1 , wherein the object detection image includes positions and names of the plurality of anatomical structures.

5 . An ultrasound image detection method based on artificial intelligence automatic labeling of an anatomical structure, the ultrasound image detection method comprising:

receiving an image to be recognized from an ultrasound apparatus by a receiving module, executed by a processor;

receiving the image to be recognized from the receiving module by an image recognition module having an object detection model and executed by the processor, such that the image recognition module uses the object detection model to detect a plurality of anatomical structures in the image to be recognized, thereby obtaining a plurality of object recognition images with object detection results;

receiving the plurality of object recognition images from the image recognition module by an image processing module executed by the processor, and using at least one of the plurality of object recognition images received first as at least one reference image and the other one of the plurality of object recognition images received subsequently as an image to be recognized by the image processing module, such that the image processing module computes a new confidence index of the plurality of anatomical structures in the image to be recognized according to the object detection results of the image to be recognized and the reference image, a decay coefficient and a distance between the reference image and the image to be recognized so as to label missed anatomical structures in the image to be recognized, thereby detecting the missed anatomical structures and outputting an object detection image; and

receiving the object detection image by a display module executed by the processor to display the object detection image in real time.

6 . The ultrasound image detection method of claim 5 , further comprising using a semi-supervised learning to train a teacher model and a student model by an automatic labeling module executed by the processor, such that the automatic labeling module uses the trained teacher model to label unlabeled plural ultrasound images to generate a plurality of training images, thereby providing for the image recognition module to train the object detection model.

7 . The ultrasound image detection method of claim 5 , further comprising using a Focus framework to perform a slicing operation on the received image to be recognized by the object detection model, such that a feature map of the image to be recognized is obtained, thereby providing for the object detection model to detect a plurality of anatomical structures in the image to be recognized.

8 . The ultrasound image detection method of claim 5 , wherein the object detection image includes positions and names of the plurality of anatomical structures.