Method and system for generating a confidence score using deep learning model

Patent metadata

Field	Value
Publication number	US-10997717-B2
Application number	US-201916263076-A
Country	US
Kind code	B2
Filing date	Jan 31, 2019
Priority date	Jan 31, 2019
Publication date	May 4, 2021
Grant date	May 4, 2021

Abstract

In a system and method for analyzing images, an input image is provided to a computer and is processed therein with a first deep learning model so as to generate an output result for the input image; and applying a second deep learning model is applied to the input image to generate an output confidence score that is indicative of the reliability of any output result from the first deep learning model for the input image.

First claim

The invention claimed is:

1. A method for processing an image, the method comprising:
receiving an input image and process the input image with a first deep learning model to generate an output result for the input image; and
applying a second deep learning model to the input image to generate an output confidence score which is indicative of the reliability of the output result from the first deep learning model for the input image, wherein applying the second deep learning model to generate an output confidence score comprises:
generating a reconstructed image; and
comparing the reconstructed image with the input image to generate the output confidence score.

2. The method of claim 1 , wherein the first deep learning model and the second deep learning model have been trained on the same training data.

3. The method of claim 1 , comprising generating the output confidence score using mean square error or structural similarity.

4. The method of claim 1 further comprising determining whether the output confidence score is within a confidence interval.

5. The method of claim 4 , further comprising outputting an alert when it is determined that the output confidence score is outside a confidence interval.

6. The method of claim 5 , further comprising outputting a high alert when it is determined that the output confidence score is outside a first confidence interval and outputting a low alert when it is determined that the output confidence score is outside a second confidence interval that is broader than the first confidence interval.

7. The method of claim 1 , further comprising applying the first deep learning model to the input image and simultaneously outputting the output confidence score and the output result.

8. The method of claim 1 , wherein the second deep learning model is an autoencoder.

9. A non-transitory, computer-readable data storage medium encoded with programming instructions that, when the storage medium is loaded into a computer, cause the computer to:
receive an input image and process the input image with a first deep learning model to generate an output result for the input image; and
apply a second deep learning model to the input image to generate an output confidence score which is indicative of the reliability of the output result from the first deep learning model for the input image
apply a third deep learning model to generate a second output confidence score by generating a reconstructed image, applying the first deep learning model to the reconstructed image, calculating a first loss value based on the input image and the output result using the reconstructed image, and generating the second output confidence score using the first loss value.

10. The non-transitory, computer-readable data storage medium of claim 9 , wherein the programming instructions, when the storage medium is loaded into the computer, cause the computer to calculate a second loss value using the reconstructed image and the input image and generate the output confidence score using both the first loss value and the second loss value.

11. An image processing system comprising
an image capture device configured to capture an image;
an image processor configured to:
receive said image from the image capture device as an input image,
process said input image with a first deep learning model to generate an output result for the input image, and
apply a second deep learning model to the input image to generate an output confidence score that is indicative of the reliability of the output result from the first deep learning model for the input image by generating a reconstructed image, applying the first deep learning model to the reconstructed image, calculating a first loss value based on the input image and the output result using the reconstructed image, and generating the output confidence score using the first loss value; and

a user interface configured to display at least one of the output confidence score and the output result that are generated by the image processor.

12. The image processing system of claim 11 , wherein the applying the second deep learning model to generate an output confidence score further comprises calculating a second loss value using the reconstructed image and the input image and generating the output confidence score using both the first loss value and the second loss value.

Assignees

• Siemens Healthcare Gmbh

Inventors

• Ceccaldi Pascal
• Mountney Peter
• Toth Daniel
• Cimen Serkan

Classifications

• G16H30/40 Primary

  for processing medical images, e.g. editing · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06T7/0012 Primary

  Biomedical image inspection · CPC title

• G06N3/045

  Combinations of networks · CPC title