Assisted detection model of breast tumor, assisted detection system thereof, and method for assisted detecting breast tumor

1. An assisted detection model of breast tumor, comprising following establishing steps:

obtaining a reference database, wherein the reference database comprises a plurality of reference breast ultrasound images;

performing an image preprocessing step, wherein the image preprocessing step is for dividing an image matrix value of each of the reference breast ultrasound images by a first normalization factor to obtain a reference value interval, and the reference value interval is between 0 and 1;

performing a feature selecting step, wherein the feature selecting step is for selecting a feature matrix according to the reference database by using an autoencoder module, and the autoencoder module comprises:

an encoder for compressing the reference value interval to obtain the feature matrix, wherein the encoder comprises a plurality of convolution layers and a plurality of pooling layers; and

a decoder for reducing the feature matrix and comparing the reduced feature matrix with the reference breast ultrasound images to confirm that the feature matrix comprises key information in each of the reference breast ultrasound images, wherein the decoder comprises a plurality of convolution layers and a plurality of upsampling layers; and

performing a classifying step, wherein the classifying step is for achieving a convergence of the feature matrix by using a deep learning classifier to obtain the assisted detection model of breast tumor;

wherein the assisted detection model of breast tumor is used to determine a breast tumor type of a subject and predict a probability of a tumor location of the subject.

2. The assisted detection model of breast tumor of claim 1 , wherein the first normalization factor is 255.

3. The assisted detection model of breast tumor of claim 1 , wherein the image preprocessing step further comprises:

trimming the reference breast ultrasound images; and

resetting the image size of the trimmed reference breast ultrasound images.

4. The assisted detection model of breast tumor of claim 1 , wherein a pooling function of the pooling layers is a max pooling.

5. The assisted detection model of breast tumor of claim 1 , wherein the deep learning classifier is a convolution neural network.

6. The assisted detection model of breast tumor of claim 1 , wherein the breast tumor type is no tumor, benign tumor or malignant tumor.

7. An assisted detection method of breast tumor, comprising:

providing the assisted detection model of breast tumor of claim 1 ;

providing a target breast ultrasound image of a subject;

dividing image matrix values of the target breast ultrasound image by a second normalization factor to obtain a target value interval; and

using the assisted detection model of breast tumor to analyze the target value interval to determine a breast tumor type of the subject and predict a probability of a tumor location of the subject.

8. The assisted detection method of breast tumor of claim 7 , wherein the second normalization factor is 255.

9. The assisted detection method of breast tumor of claim 7 , wherein the breast tumor type is no tumor, benign tumor or malignant tumor.

10. An assisted detection system of breast tumor, comprising:

an image capturing unit for obtaining a target breast ultrasound image of a subject; and

a non-transitory machine readable medium storing a program, which when executed by at least one processing unit, determines a breast tumor type of the subject and predicts a probability of a tumor location of the subject, the program comprising:

a reference database obtaining module for obtaining a reference database, wherein the reference database comprises a plurality of reference breast ultrasound images;

a first image preprocessing module for normalizing an image matrix value of each of the reference breast ultrasound images to obtain a reference value interval, wherein the reference value interval is between 0 and 1;

an autoencoder module for selecting a feature matrix according to the reference database, the autoencoder module comprising:

an encoder for compressing the reference value interval to obtain the feature matrix, wherein the encoder comprises a plurality of convolution layers and a plurality of pooling layers; and

a decoder for reducing the feature matrix and comparing the reduced feature matrix with the reference breast ultrasound images to confirm that the feature matrix comprises key information in each of the reference breast ultrasound images, wherein the decoder comprises a plurality of convolution layers and a plurality of upsampling layers;

a classifying module for achieving a convergence of the feature matrix by using a deep learning classifier to obtain an assisted detection model of breast tumor;

a second image preprocessing module for normalizing an image matrix value of each of the target breast ultrasound image to obtain a target value interval, wherein the target value interval is between 0 and 1; and

a comparing module for analyzing the target value interval by the assisted detection model of breast tumor to determine the breast tumor type of the subject and predict the probability of the tumor location of the subject.

11. The assisted detection system of breast tumor of claim 10 , wherein the first image preprocessing module comprises sets of instructions for:

trimming the reference breast ultrasound images;

dividing an image matrix value of each of the reference breast ultrasound images by a first normalization factor to obtain a reference value interval; and

resetting the image size of the trimmed reference breast ultrasound images.

12. The assisted detection system of breast tumor of claim 11 , wherein the first normalization factor is 255.

13. The assisted detection system of breast tumor of claim 10 , wherein a pooling function of the pooling layers is a max pooling.

14. The assisted detection system of breast tumor of claim 10 , wherein the deep learning classifier is a convolutional neural network.

15. The assisted detection system of breast tumor of claim 10 , wherein the second image preprocessing module comprises sets of instructions for:

trimming the target breast ultrasound image;

dividing an image matrix value of the target breast ultrasound image by a second normalization factor to obtain a target value interval; and

resetting the image size of the trimmed target breast ultrasound image.

16. The assisted detection system of breast tumor of claim 15 , wherein the second normalization factor is 255.

17. The assisted detection system of breast tumor of claim 10 , wherein the breast tumor type is no tumor, benign tumor or malignant tumor.