Medical image analyzing system and method thereof

Patent metadata

Field	Value
Publication number	US-11684333-B2
Application number	US-202017137647-A
Country	US
Kind code	B2
Filing date	Dec 30, 2020
Priority date	Jan 3, 2020
Publication date	Jun 27, 2023
Grant date	Jun 27, 2023

Abstract

Provided is a medical image analyzing system and a method thereof, which includes: acquiring a processed image having a segmentation label corresponding to a cancerous part of an organ (if present), generating a plurality of image patches therefrom, performing feature analysis on the image patches and model training to obtain prediction values, drawing a receiver operating characteristic curve using the prediction values, and determining a threshold with which determines whether the image patches are cancerous, so as to effectively improve the detection rate of, for example, pancreatic cancer.

First claim

What is claimed is:

1. A medical image analyzing system, comprising:
an image preprocessing module, executed by a processor, configured to process at least one image corresponding to an organ to generate at least one processed image, wherein the processed image comprises a segmentation label corresponding to a cancerous part of the organ;
a patch cropping module, executed by the processor, configured to acquire a plurality of image patches from the processed image;
a feature analysis module, executed by the processor, configured to perform feature analysis on the plurality of image patches to obtain a plurality of feature values corresponding to each of the plurality of image patches;
a training module, executed by the processor, configured to train a full model using the plurality of feature values of each of the plurality of image patches to obtain a plurality of first prediction values corresponding to the respective plurality of image patches;
a threshold selection module, executed by the processor, configured to plot a first curve based on the plurality of first prediction values and determine a first threshold that is used for determining whether each of the plurality of image patches is cancerous according to the first curve; and
a computer-assisted detection/diagnosis module, executed by the processor, configured to input at least one patient image to the image preprocessing module and the patch cropping module to generate a plurality of patient image patches and input the plurality of patient image patches into the full model to obtain a plurality of first prediction values corresponding to the respective plurality of patient image patches,
wherein the computer-assisted detection/diagnosis module further enables the threshold selection module to calculate at least one second prediction value corresponding to the at least one patient image based on the plurality of first prediction values respectively corresponding to the plurality of patient image patches, and plot a second curve based on the at least one second prediction value to determine a second threshold that is used for determining whether the at least one patient image is cancerous according to the second curve.

2. The medical image analyzing system of claim 1 , wherein the patch cropping module acquires the plurality of image patches with a square window that moves along an x axis and a y axis of the processed image.

3. The medical image analyzing system of claim 2 , wherein the patch cropping module masks the segmentation label of the processed image and masks the processed image except for the organ, and then acquires an unmasked area of the processed image with the square window moving at a stride of 5 pixels to generate the plurality of image patches.

4. The medical image analyzing system of claim 2 , wherein the patch cropping module masks the processed image except for the segmentation label, and then acquires an unmasked area of the processed image with the square window moving at a stride of 1 pixel to generate the plurality of image patches.

5. The medical image analyzing system of claim 1 , wherein the feature analysis module performs feature analysis using radiomics.

6. The medical image analyzing system of claim 5 , wherein features employed by the radiomics include: First Order features, Gray Level Co-occurrence Matrix (GLCM) features, Gray Level Dependence Matrix (GLDM) features, Gray Level Run Length Matrix (GLRLM) features, Gray Level Size Zone Matrix (GLSZM) features or Neighboring Gray Tone Difference Matrix (NGTDM) features.

7. The medical image analyzing system of claim 1 , wherein the training module trains the full model using an extreme gradient boosting (XGboost) machine learning algorithm.

8. The medical image analyzing system of claim 1 , wherein the at least one second prediction value is a ratio between a number of patient image patches that are predicted as cancerous in the at least one patient image by applying the first threshold to the plurality of first prediction values respectively corresponding to each of the plurality of patient image patches and a total number of the plurality of patient image patches.

9. The medical image analyzing system of claim 1 , further comprising a feature selection module, executed by the processor, configured to generate a feature importance ranking based on a plurality of features corresponding to the plurality of feature values in the full model, and enable the training module to train a reduced model using the feature value of at least one of the plurality of features of each of the plurality of image patches to obtain a plurality of first reduced prediction values corresponding to the respective plurality of image patches, and enable the threshold selection module to plot a first reduced curve based on the plurality of first reduced prediction values.

10. The medical image analyzing system of claim 9 , wherein the feature importance ranking is ranked according to numbers of occurrences of the features, gain values of the features, or any combination thereof.

11. The medical image analyzing system of claim 9 , wherein the feature selection module enables the training module to train the reduced model using the feature value of a feature starting from a top of the feature importance ranking.

12. The medical image analyzing system of claim 9 , wherein the computer-assisted detection/diagnosis module further inputs the plurality of patient image patches into the reduced model to obtain a plurality of first reduced prediction values respectively corresponding to the plurality of patient image patches, and enables the threshold selection module to calculate at least one second reduced prediction value corresponding to the at least one patient image based on the plurality of first reduced prediction values respectively corresponding to the plurality of patient image patches, and plot a second reduced curve based on the at least one second reduced prediction value.

13. The medical image analyzing system of claim 12 , wherein if an area under receiver operating characteristic curve (AUC) of the first reduced curve does not equate to or approximate an AUC of the first curve, or an AUC of the second reduced curve does not equate to or approximate an AUC of the second curve, the feature selection module enables the training module to train the reduced model using a feature value of the feature that is ranked first in the feature importance ranking and a feature value of a subsequent feature that is ranked next in the feature importance ranking iteratively until the AUCs of the first reduced curve and the second reduced curve equate to or approximate the AUCs of the first curve and the second curve, respectively.

14. A medical image analyzing method, comprising:
processing at least one image corresponding to an organ to generate at least one processed image, wherein the processed image comprises a segmentation label corresponding to a cancerous part of the organ;
acquiring a plurality of image patches from the processed image;
performing feature analysis on the plurality of image patches to obtain a plurality of feature values corresponding to each of the plurality of image patches;
training a full model using the plurality of feature values of each of the plurality of image patches to obtain a plurality of first prediction values corresponding to the respective plurality of image patches;
plotting a first curve based on the plurality of first prediction values and determining a first threshold that is used for determining whether each of the plurality of image patches is cancerous according to the first curve;
inputtin

Assignees

• Univ Nat Taiwan

Inventors

• Liao Wei-Chih
• Wang Wei-Chung
• Liu Kao-Lang
• Chen Po-Ting
• Chang da-wei

Classifications

• G16H50/20

  for computer-aided diagnosis, e.g. based on medical expert systems · CPC title

• A61B5/055

  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/30096

  Tumor; Lesion · CPC title

• G06T2207/10088

  Magnetic resonance imaging [MRI] · CPC title