Evaluating input data using a deep learning algorithm

Patent metadata

Field	Value
Publication number	US-12412092-B2
Application number	US-202318216629-A
Country	US
Kind code	B2
Filing date	Jun 30, 2023
Priority date	Sep 20, 2017
Publication date	Sep 9, 2025
Grant date	Sep 9, 2025

Abstract

The invention provides a method for evaluating a set of input data, the input data comprising at least one of: clinical data of a subject; genomic data of a subject; clinical data of a plurality of subjects; and genomic data of a plurality of subjects, using a deep learning algorithm. The method includes obtaining a set of input data, wherein the set of input data comprises raw data arranged into a plurality of data clusters and tuning the deep learning algorithm based on the plurality of data clusters. The deep learning algorithm comprises: an input layer; an output layer; and a plurality of hidden layers. The method further includes performing statistical clustering on the raw data using the deep learning algorithm, thereby generating statistical clusters and obtaining a marker from each statistical cluster. Finally, the set of input data is evaluated based on the markers to derive data of medical relevance in respect of the subject or subjects.

First claim

The invention claimed is:

1. A computer-implemented method for tuning a deep learning neural network (DLNN) trained using input data comprising clinical data and/or genomic data, the method comprising:
performing clustering on the input data using the DLNN to generate a plurality of clusters by identifying principal variables of hidden layers of the DLNN so as to reduce a number of computations required, and considering hidden layers of the DLNN as low dimensional representations;
identifying data points that are in different clusters of the plurality of clusters, wherein the identified data points are in: (i) a first hidden layer of the hidden layers of the DLNN and (ii) the same cluster of the plurality of clusters in a second hidden layer of the hidden layers of the DLNN;
identifying a borderline case as including the identified data points based on the data points being identified as being in different clusters in the first hidden layer and in the same cluster in the second hidden layer;
providing the borderline case to a display;
receiving feedback from a user regarding whether the identified data points belong in the same cluster; and
updating the DLNN based on the feedback received.

2. The method of claim 1 , further comprising:
determining a Gaussian mean width of the input data;
determining a convergence rate of a loss-function of the DLNN; and
selecting hidden layers based on the Gaussian mean width and the convergence rate.

3. The method of claim 2 , wherein the determining of the Gaussian mean width is based on the size of the plurality of clusters.

4. The method of claim 2 , wherein the determining of the Gaussian mean width is based on the number of the plurality of clusters.

5. A non-transitory computer readable medium having stored thereon a computer program code, wherein such computer program code when run on a computer, causes the computer to execute the method of claim 1 .

6. A method as claimed in claim 1 , further comprising predicting a survival rate of a subject using the tuned DLNN.

7. A non-transitory computer readable medium having stored thereon a computer program code, wherein such computer program code when run on a computer, causes the computer to execute the method of claim 6 .

8. The method of claim 1 , further comprising identifying a plurality of borderline cases, wherein a threshold to an angle is used to limit the number of borderline cases provided to the user.

9. The method of claim 1 , wherein the clustering includes k-means clustering.

10. The method of claim 1 , wherein the clustering includes clustering between adjacent hidden layers of the DLNN.

11. The method of claim 1 , further comprising:
obtaining respective markers from the plurality of clusters, wherein the markers are biomarkers respectively related to clinical parameters respectively contained within the clusters.

12. The method of claim 11 , further comprising:
evaluating a set of the input data based on the markers to derive data of medical relevance in respect of a subject or subjects, wherein the markers are evaluated with reference to historical subject data collected from subjects with similar conditions and/or symptoms to determine an effective treatment method.

13. The method of claim 1 , wherein the DLNN is an autoencoder.

14. A system for tuning a deep learning neural network (DLNN) trained using input data comprising clinical data and/or genomic data, the system comprising:
a deep learning neural network (DLNN) trained using input data comprising clinical data and/or genomic data;
a display; and
a processor configured to: (i) perform clustering on the input data using the DLNN to generate a plurality of clusters by identifying principal variables of hidden layers of the DLNN so as to reduce a number of computations required, and considering hidden layers of the DLNN as low dimensional representations; (ii) identify data points that are in different clusters of the plurality of clusters, wherein the identified data points are in: (1) a first hidden layer of the hidden layers of the DLNN and (2) the same cluster of the plurality of clusters in a second hidden layer of the hidden layers of the DLNN; (iii) identify a borderline case as including the identified data points based on the data points being identified as being in different clusters in the first hidden layer and in the same cluster in the second hidden layer; (iv) provide the borderline case to a display; (v) receive feedback from a user regarding whether the identified data points belong in the same cluster; and (vi) update the DLNN based on the feedback received.

15. The system of claim 14 , wherein the processor is further configured to:
determine a Gaussian mean width of the input data;
determine a convergence rate of a loss-function of the DLNN; and
select hidden layers based on the Gaussian mean width and the convergence rate.

16. The system of claim 15 , wherein determining the Gaussian mean width is based on a size of the plurality of clusters.

17. The system of claim 15 , wherein determining the Gaussian mean width is based on a number of the plurality of clusters.

18. The system of claim 14 , further comprising predicting a survival rate of a subject using the tuned DLNN.

19. The system of claim 14 , wherein the processor is configured to identify a plurality of borderline cases, wherein a threshold to an angle is used to limit the number of borderline cases provided to the user.

20. The system of claim 14 , wherein clustering comprises k-means clustering.

21. The system of claim 14 , wherein clustering comprises clustering between adjacent hidden layers of the DLNN.

Assignees

• Koninklijke Philips Nv

Inventors

• Mavroeidis Dimitrios
• Hendriks Monique
• Vos Pieter Christiaan
• Consoli Sergio
• Kustra Jacek Lukasz
• Janssen Johan
• Hoffmann Ralf Dieter

Classifications

• G06N3/0455

  Auto-encoder networks; Encoder-decoder networks · CPC title

• G06N3/091

  Active learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06F18/23213

  with fixed number of clusters, e.g. K-means clustering · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title