Computer-aided diagnostic system for early diagnosis of prostate cancer

1. A method for diagnosing prostate cancer comprising:

obtaining magnetic resonance imaging prostate data at a plurality of b-values;

processing the magnetic resonance imaging prostate data through a plurality of first autoencoders to create a plurality of imaging output probability data, wherein each of the plurality of first autoencoders processes the magnetic resonance imaging prostate data associated with only one of the plurality of b-values;

organizing the plurality of imaging output probability data into an input vector, wherein the input vector is a one-dimensional list of the imaging output probability data;

obtaining biological data from a biological value test, wherein the biological data is not the magnetic resonance imaging prostate data;

processing the biological data through a data classifier to create biological output probability data, wherein the input vector further includes the biological output probability data;

processing the input vector through a second autoencoder to generate a diagnostic probability of prostate cancer; and

generating a diagnosis based on the diagnostic probability.

2. The method of claim 1 , wherein the biological value test is based at least in part on a test selected from the group consisting of a digital rectal examination, a prostate specific antigen blood test, a needle biopsy, a patient questionnaire, or a histology-based test.

3. The method of claim 1 , wherein the data classifier is based at least in part on a low-dimensional classifier selected from the group consisting of a K-nearest neighbor classifier or a softmax regression classifier.

4. The method of claim 1 , wherein the plurality of first autoencoders and the second autoencoder are selected from the group consisting of a stacked non-negativity constrained sparse autoencoder, a non-negatively constrained sparse autoencoder, a softmax regression classifier, or a convolutional neural network.

5. The method of claim 1 , wherein obtaining magnetic resonance imaging prostate data comprises:

determining an intensity-based imaging feature from a magnetic resonance imaging scan;

creating a plurality of intensity-based feature maps from magnetic resonance imaging data based on the plurality of b-values, wherein each of the plurality of intensity-based feature maps is associated with only one of the plurality of b-values; and

generating magnetic resonance imaging prostate data by applying the plurality of intensity-based feature maps to a plurality of cumulative distribution functions based on the plurality of b-values, wherein each of the plurality of cumulative distribution functions is associated with only one of the plurality of b-values.

6. The method of claim 5 , wherein the magnetic resonance imaging scan is a diffusion-weighted magnetic resonance imaging scan.

7. The method of claim 5 , wherein the intensity-based imaging feature is an apparent diffusion coefficient.

8. The method of claim 1 , wherein the plurality of b-values are selected from the range of about 0 s/mm 2 to about 700 s/mm 2 .

9. A system, comprising:

one or more processors; and

memory containing program code that, when executed by at least one of the one or more processors, is configured to:

process magnetic resonance imaging prostate data through a plurality of first autoencoders to create a plurality of imaging output probability data, wherein each of the plurality of first autoencoders processes the magnetic resonance imaging prostate data associated with only one of a plurality of b-values;

organize the plurality of imaging output probability data into an input vector, wherein the input vector is a one-dimensional list of the imaging output probability data;

process biological data through a data classifier, wherein the biological data is not the magnetic resonance imaging prostate data, and wherein the input vector further includes the biological output probability data;

process the input vector through a second autoencoder to generate a diagnostic probability of prostate cancer; and

generate a diagnosis based on the diagnostic probability.

10. The system of claim 9 , wherein the data classifier is based at least in part on a low-dimensional classifier selected from the group consisting of a K-nearest neighbor classifier or a softmax regression classifier.

11. The system of claim 9 , wherein the plurality of first autoencoders and the second autoencoder are selected from the group consisting of a stacked non-negativity constrained sparse autoencoder, a non-negatively constrained sparse autoencoder, a softmax regression classifier, or a convolutional neural network.

12. The system of claim 9 , wherein obtaining magnetic resonance imaging prostate data further causes the system to:

determine an intensity-based imaging feature from a magnetic resonance imaging scan;

create a plurality of intensity-based feature maps from magnetic resonance imaging data based on the plurality of b-values, wherein each of the plurality of intensity-based feature maps is associated with only one of the plurality of b-values; and

generate magnetic resonance imaging prostate data by applying the plurality of intensity-based feature maps to a plurality of cumulative distribution functions based on the plurality of b-values, wherein each of the plurality of cumulative distribution functions is associated with only one of the plurality of b-values.

13. The system of claim 12 , wherein the magnetic resonance imaging scan is a diffusion-weighted magnetic resonance imaging scan.

14. The system of claim 12 , wherein the intensity-based imaging feature is an apparent diffusion coefficient.

15. The system of claim 9 , wherein the plurality of b-values are selected from the range of about 0 s/mm 2 to about 700 s/mm 2 .