System and method for detecting cancerous tissue from a thermal image
What is disclosed is a system and method for the detection of cancerous tissue by analyzing blocks of pixels in a thermal image of a region of exposed skin tissue. In one embodiment, matrices are received which have been derived from vectors of temperature values associated with pixels in blocks of pixels which have been isolated from a plurality of thermal images of both cancerous and non-cancerous tissue. The vectors are rearranged to form matrices. A thermal image of a subject is received. Blocks of pixels which reside within a region of exposed skin tissue are identified and isolated. For each identified pixel block, an image vector comprising temperature values associated with these pixels is formed. The vector is provided to a classifier which uses the matrices to classify tissue associated with this block of pixels as being either cancerous or non-cancerous tissue.
1. A method for detecting cancerous tissue from a thermal image, the method comprising:
receiving, with a processor, matrices A H of size m×H and A N of size m×N comprising vectors of size m×1 derived from temperature values associated with pixels in blocks of size p×q, where m=pq, which have been isolated from a plurality of thermal images of cancerous and non-cancerous tissues, where H and N represent a number of cancerous and non-cancerous blocks of pixels, respectively, said vectors being rearranged to form said matrices;
receiving a thermal image of a subject;
identifying a block of pixels in a region of exposed skin tissue in said received thermal image;
forming a vector y of size m×1 derived from temperature values associated with pixels in said identified block of pixels; and
providing said vector to a classifier, said classifier using said matrices to classify tissue in said thermal image represented by said block of pixels as being one of: cancerous and a non-cancerous tissue.
2. The method of claim 1 , wherein vector y is represented by:
y=[A H A N ]x
where x=[x H T x N T ] T is a coefficient vector of size (N+H)×1, x H is a vector of size H×1 and x N is a vector of size N×1 obtained from blocks of pixels from said plurality of thermal images of cancerous and non-cancerous tissues, respectively, and T is a transpose operation.
3. The method of claim 2 , further comprising:
performing minimization using vector y and matrices A H and A N to obtain a sparse coefficient vector x*, where x*=[(x H *) T (x N *) T ] T ; and
determining residues represented by:
R N =∥y−A N x N *∥ 2 ; and
R H =∥y−A H x H *∥ 2
where R H and R N are residues of cancerous and non-cancerous tissues, respectively.
4. The method of claim 3 , wherein said minimization is a constrained minimization with convex optimization.
5. The method of claim 3 , wherein, in response to R H <R N , classifying tissue in said block of pixels as cancerous.
6. The method of claim 3 , wherein, in response to
log
(
R
N
R
H
)
>
0
,
classifying tissue in said block of pixels as cancerous.
7. The method of claim 1 , wherein said classifier is any of: Support Vector Machine (SVM), a neural network, a Bayesian network, a Logistic regression, Naïve Bayes, Randomized Forests, Decision Trees, Boosted Decision Trees, K-nearest neighbor, Restricted Boltzmann Machine (RBM), and a hybrid system comprising any combination hereof.
8. The method of claim 1 , wherein, in advance of identifying a block of pixels in said thermal image for processing, further comprising enhancing a spatial resolution of said thermal image.
9. The method of claim 1 , wherein said matrices A H and A N are one of; basis matrices, product of sampling matrices and basis matrices, dictionary matrices, and random matrices.
10. The method claim 1 , wherein said vector y comprises temperature values containing a number of rows which is less than a number of rows in matrices A H and A N .
11. The method of claim 1 , wherein, in response to said subject being classified as having cancerous tissue, performing any of: initiating an alert, and signaling a medical professional.
12. The method of claim 1 , further comprising communicating said subject's classification to any of: a memory, a storage device, a display device, a handheld wireless device, a handheld cellular device, and a remote device over a network.
13. A system for detecting cancerous tissue from a thermal image, the system comprising:
a classifier; and
a processor in communication with a memory and said classifier, said processor executing machine readable instructions for performing:
receiving matrices A H of size m×H and A N of size m×N comprising vectors of size m×1 derived from temperature values associated with pixels in blocks of size p×q, where m=pq, which have been isolated from a plurality of thermal images of cancerous and non-cancerous tissues, where H and N represent a number of cancerous and non-cancerous blocks of pixels, respectively, said vectors being rearranged to form said matrices;
receiving a thermal image of a subject;
identifying a block of pixels in a region of exposed skin tissue in said received thermal image;
forming a vector y of size m×1 derived from temperature values associated with pixels in said identified block of pixels; and
providing said vector to said classifier, said classifier using said matrices to classify tissue in said thermal image represented by said block of pixels as being one of: cancerous and a non-cancerous tissue.
14. The system of claim 13 , wherein said vector y is represented by:
y=[A H A N ]x
where x=[x H T x N T ] T is a coefficient vector of size (N+H)×1, x H is a vector of size H×1 and x N is a vector of size N×1 obtained from blocks of pixels from said plurality of thermal images of cancerous and non-cancerous tissues, respectively, and T is a transpose operation.
15. The system of claim 14 , further comprising:
performing minimization using vector y and matrices A H and A N to obtain a sparse coefficient vector x*, where x*=[(x H *) T (x N *) T ] T ; and
determining residues represented by:
R N =∥y−A N x N *∥ 2 ; and
R H =∥y−A H x H *∥ 2
where R H and R N are residues of cancerous and non-cancerous tissues, respectively.
16. The system of claim 15 , wherein said minimization is a constrained minimization with convex optimization.
17. The system of claim 15 , wherein, in response to R H <R N , classifying tissue in said block of pixels as cancerous.
18. The system of claim 15 , wherein, in response to
log
(
R
N
R
H
)
>
0
,
classifying tissue in said block of pixels as cancerous.
19. The system of claim 13 , wherein said classifier is any of: Support Vector Machine (SVM), a neural network, a Bayesian network, a Logistic regression, Naïve Bayes, Randomized Forests, Decision Trees, Boosted Decision Trees, K-nearest neighbor, Restricted Boltzmann Machine (RBM), and a hybrid system comprising any combination hereof.
20. The system of claim 13 , wherein, in advance of identifying a block of pixels in said thermal image for processing, further comprising enhancing a spatial resolution of said thermal image.
21. The system of claim 13 , wherein said matrices A H and A N are one of; basis matrices, product of sampling matrices and basis matrices, dictionary matrices, and random matrices.
22. The system of claim 13 , wherein said vector y comprises temperature values containing a number of rows which is less than a number of rows in matrices A H and A N .
23. The system of claim 13 , wherein, in response to said subject being classified as having cancerous tissue, performing any of: initiating an alert, and signaling a medical professional.
24. The system of claim 13 , further comprising communicating said subject's classification to any of: a memory, a storage device, a display device, a handheld wireless device, a handheld cellular device, and a remote device over a network.
25. The system of claim 13 , wherein said classification occurs in real-time.