Computer-based diagnostic system

Patent metadata

Field	Value
Publication number	US-10910094-B2
Application number	US-201816196387-A
Country	US
Kind code	B2
Filing date	Nov 20, 2018
Priority date	Nov 24, 2017
Publication date	Feb 2, 2021
Grant date	Feb 2, 2021

Abstract

A computer-based diagnostic system includes an image generation unit adapted to generate tomographic images of a patient's organ; a deep machine learning unit configured to process generated tomographic images of the patient's organ to classify organ regions of diseased functional tissue of the patient's organ as belonging to one of a set of abnormal image patterns using trained deep neural networks; and a clinical decision support unit adapted to process classification results of the deep machine learning unit to calculate a diagnostic result output via an interface of the clinical decision support unit.

First claim

What is claimed is:

1. A computer-based diagnostic system, comprising:
a deep machine learning unit configured to classify a region of an organ of a patient as corresponding to one of a set of abnormal image patterns based on a tomographic image of the organ using a plurality of trained convolutional neural networks (CNNs) to obtain a classification result, the set of abnormal image patterns corresponding to diseased function tissue; and
processing circuitry configured to
calculate a diagnostic result based on the classification result, and
output the diagnostic result via an interface,

wherein
the plurality of trained CNNs correspond to a plurality of processing stages of the deep machine learning unit,
the plurality of trained CNNs includes a first trained CNN corresponding to a localization stage among the plurality of processing stages, the first trained CNN configured to delineate a plurality of respective regions of diseased functional tissue from regions of healthy tissue within a segment of the organ, the plurality of respective regions including the region, and
the plurality of trained CNNs includes a second trained CNN corresponding to a classification stage among the plurality of processing stages, the second trained CNN configured to classify each of the plurality of respective regions of diseased functional tissue as corresponding to a respective one of the set of abnormal image patterns.

2. The computer-based diagnostic system of claim 1 , wherein the plurality of trained CNNs includes a third trained CNN corresponding to a preprocessing stage among the plurality of processing stages, the third trained CNN configured to provide a segment of the organ from the tomographic image.

3. The computer-based diagnostic system of claim 1 , wherein the first trained CNN is a constrained convolutional neural network (CCNN).

4. The computer-based diagnostic system of claim 1 , wherein the processing circuitry is configured to calculate the diagnostic result based on the classification result and additional clinical parameters of the patient.

5. The computer-based diagnostic system of claim 1 , wherein
the tomographic image is one in a sequence of tomographic images of the organ captured over time;
the deep machine learning unit is configured to classify each tomographic image in the sequence of tomographic images to obtain a plurality of classification results, the plurality of classification results including the classification result; and
the processing circuitry is configured to calculate the diagnostic result based on the plurality of classification results, the diagnostic result indicating at least one of a change, a progression, a rate of change, a prognosis or a risk classification.

6. The computer-based diagnostic system of claim 1 , wherein the tomographic image is one of a plurality of tomographic images, each of the plurality of tomographic images corresponding to an image slice of the organ, and each of the plurality of tomographic images including a plurality of image pixels.

7. The computer-based diagnostic system of claim 1 , wherein the deep machine learning unit is configured to calculate scores for different abnormal image patterns among the set of abnormal image patterns to which the region of the organ corresponds during the classification stage.

8. The computer-based diagnostic system of claim 7 , wherein the scores for the different abnormal image patterns include a percentage score of a respective abnormal image pattern computed based on a ratio between
a number of voxels of the tomographic image within the region of the organ and classified as corresponding to the respective abnormal image pattern, and
a total number of voxels in the region of the organ.

9. The computer-based diagnostic system of claim 1 , wherein the tomographic image is of a chest of the patient; and the organ is a lung organ.

10. The computer-based diagnostic system of claim 9 , wherein the deep machine learning unit is configured to provide a segmentation mask for all lung lobs of the lung organ during a preprocessing stage.

11. The computer-based diagnostic system of claim 10 , wherein the set of abnormal image patterns includes image patterns corresponding to fibrosis, emphysema, honeycombing, ground glass opacity, reticular, cysts, consolidation and mosaic pattern attenuation.

12. The computer-based diagnostic system of claim 10 , wherein the diagnostic result includes an indication of whether the region of the lung organ is affected by at least one of idiopathic pulmonary fibrosis (IFP) or related interstitial lung disease (ILD).

13. The computer-based diagnostic system of claim 9 , wherein the set of abnormal image patterns includes image patterns corresponding to fibrosis, emphysema, honeycombing, ground glass opacity, reticular, cysts, consolidation and mosaic pattern attenuation.

14. The computer-based diagnostic system of claim 9 , wherein the diagnostic result includes an indication of whether the region of the lung organ is affected by at least one of idiopathic pulmonary fibrosis (IFP) or interstitial lung disease (ILD).

15. A computer-based diagnostic method, comprising:
classifying, by a deep machine learning unit, a region of an organ of a patient as corresponding to one of a set of abnormal image patterns based on a tomographic image of the organ using a plurality of trained convolutional neural networks (CNNs) to obtain a classification result, the set of abnormal image patterns corresponding to diseased function tissue;
calculating a diagnostic result based on the classification result; and
outputting the diagnostic result via an interface, wherein
the plurality of trained CNNs correspond to a plurality of processing stages of the deep machine learning unit, the plurality of trained CNNs includes a first trained CNN corresponding to a localization stage among the plurality of processing stages and a second trained CNN corresponding to a classification stage among the plurality of processing stages, and
the classifying includes
delineating, by the first trained CNN, a plurality of respective regions of diseased functional tissue from regions of healthy tissue within a segment of the organ, the plurality of respective regions including the region, and
classifying, by the second trained CNN, each of the plurality of respective regions of diseased functional tissue as corresponding to a respective one of the set of abnormal image patterns.

16. A non-transitory computer-readable medium storing instructions that, when executed by processing circuitry, cause the processing circuitry to perform a computer-based diagnostic method, the method comprising:
classifying, by a deep machine learning unit, a region of an organ of a patient as corresponding to one of a set of abnormal image patterns based on a tomographic image of the organ using a plurality of trained convolutional neural networks (CNNs) to obtain a classification result, the set of abnormal image patterns corresponding to diseased function tissue;
calculating a diagnostic result based on the classification result; and
outputting the diagnostic result via an interface, wherein
the plurality of trained CNNs correspond to a plurality of processing stages of the deep machine learning unit, the plurality of trained CNNs includes a first trained CNN corresponding to a localization stage among the plurality of processing stages and a second trained CNN corresponding to a classification stage among the plurality of processing stages, and
the classifying includes
delineating, by the first trained CNN, a plurality of

Assignees

• Siemens Healthcare Gmbh

Inventors

• Hartung Andre
• Ionasec Razvan

Classifications

• G06V10/82

  using neural networks · CPC title

• G06V10/764

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

• G16H30/40 Primary

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

• G16H30/20 Primary

  for handling medical images, e.g. DICOM, HL7 or PACS · CPC title

• G06F18/241

  relating to the classification model, e.g. parametric or non-parametric approaches · CPC title