Method and system for determining the risk of occurrence of prostate cancer
View Patent ↗Clinical information, molecular information and/or computer-generated morphometric information is used in a predictive model for predicting the occurrence of a medical condition. In an embodiment, a model predicts risk of prostate cancer progression in a patient, where the model is based on features including one or more (e.g., all) of preoperative PSA, dominant Gleason Grade, Gleason Score, at least one of a measurement of expression of AR in epithelial and stromal nuclei and a measurement of expression of Ki67-positive epithelial nuclei, a morphometric measurement of average edge length in the minimum spanning tree (MST) of epithelial nuclei, and a morphometric measurement of area of non-lumen associated epithelial cells relative to total tumor area. In some embodiments, the morphometric information is based on image analysis of tissue subject to multiplex immunofluorescence and may include characteristic(s) of a minimum spanning tree (MST) and/or a fractal dimension observed in the images.
1. A system for determining the risk of occurrence of prostate cancer in a patient, the system comprising:
(1) a fluorescence imaging device configured to record at least one sample image of a tissue sample of a patient, the tissue sample treated with a plurality of fluorochrome labeled antibodies, wherein the antibodies are selected to bind with at least AR and Ki67;
(2) a database configured to store patient data including clinical feature data values for the patient comprising a value indicative of the biopsy Gleason score (bGS) and biopsy Gleason grade (bGG) of the patient and a value indicative of the level of Prostate Specific Antigen (PSA) in the blood of the patient;
(3) a processor configured by code executing therein to perform the following:
a. evaluate the at least one sample image recorded by the imaging device and generate:
(i) one or more molecular feature values indicative of a combined AR dynamic range where the measured bGG values is <=3, and a total Ki67 where the measured bGG value is 4>=by applying at least a segmentation analysis to the sample image using a quad-tree function to differentiate the sample image into background and non-background objects where the background objects have an average pixel intensity below a predetermined threshold; and
(ii) a plurality of morphometric measurements, including: (a) a mean distance between epithelial tumor cells (MST) where the bGG value is 3 or the actual Gleason grade when bGG is 4 or higher, and (b) an area of isolated (non-lumen associated) tumor epithelial cells relative to tumor area value;
b. combine the molecular feature values, morphometric measurements and clinical feature data values into a patient database;
c. evaluate the patient data with a Support Vector Regression for Censored Data (SVRc) algorithm executed as code by the processor, the SVRc algorithm configured to output a value corresponding to a risk score for cancer occurrence based on the patient data,
wherein the SVRc algorithm is generated by performing regression, using code executed in the processor, on patient entries in a censored and uncensored patient database, where each patient entry in the censored and uncensored patient database includes data corresponding to the clinical feature data values, molecular feature values and morphometric measurements of the patient data, and where the population includes members where a cancer occurrence status is known (uncensored members) and members where a cancer occurrence status is unknown (censored members), and the regression includes implementing, as code executed in the processor, a first loss function on the censored member data such that:
Loss
(
f
(
x
)
,
y
,
s
=
1
)
=
{
C
s
*
(
e
-
ɛ
s
*
)
e
>
ɛ
s
*
0
-
ɛ
s
≤
e
≤
ɛ
s
*
C
s
(
ɛ
s
-
e
)
e
<
-
ɛ
s
,
where e=ƒ(x)−y; and ∈ and C values differentiate the penalties incurred for f(x)>0 versus f(x)<0; and a second loss function on the uncensored member data such that:
Loss
(
f
(
x
)
,
y
,
s
=
0
)
=
{
C
n
*
(
e
-
ɛ
n
*
)
e
>
ɛ
0
-
ɛ
n
≤
e
≤
ɛ
n
*
,
C
n
(
ɛ
n
-
e
)
e
<
-
ɛ
n
,
where e=ƒ(x)−y and ∈ n *≦∈ n and C n *≧C n and where ƒ(x) is the predicted time to event for sample x, s is the censorship value, and y is a target value corresponding to either the actual time to cancer reoccurrence for an entry in the censored and uncensored patient database where a cancer occurrence status is uncensored, or the last known observation time for the entry in the censored and uncensored patient database where the cancer occurrence status is censored;
d. assigns the patient to a low risk of cancer occurrence category where the output value is below 30 and assigns the patient to a high risk category where the output value is above 30; and
e. generates a report based on the updated patient database.
2. A method for determining the risk of occurrence of prostate cancer in a patient, the method comprising:
(1) capturing, with a fluorescence imaging device, at least one sample image of a tissue sample treated with a plurality of fluorochrome labeled antibodies, wherein the antibodies are selected to bind with at least AR and Ki67;
(2) accessing from a database, using a processor configured by code executing therein, a patient dataset containing clinical feature data values for the patient corresponding to at least a value indicative of the biopsy Gleason score (bGS) and biopsy Gleason grade (bGG) of the patient obtained from a tissue sample of the patient and a value indicative of the level of Prostate Specific Antigen (PSA) in the blood of the patient;
(3) evaluating, using the processor, the at least one sample image recorded by the imaging device and deriving:
a. one or more molecular feature values indicative of a combined AR dynamic range where the measured bGG values is <=3, and a total Ki67 where the measured bGG value is 4>=by applying at least a segmentation analysis to the sample image using a quad-tree function to differentiate the sample image into background and non-background objects where the background objects have an average intensity below a predetermined threshold; and
b. a plurality of morphometric measurements, including: (a) a mean distance between epithelial tumor cells where the bGG value is <=3, and (b) an area of isolated (non-lumen associated) tumor epithelial cells relative to tumor area value;
(4) updating the patient dataset using the processor, by associating with each patient dataset the derived molecular feature values and morphometric measurements;
(5) evaluating, using the processor, the updated patient dataset of step (4) with a Support Vector Regression for Censored Data (SVRc) algorithm configured as code executing in the processor, where the SVRc algorithm configures the processor to output a value corresponding to a risk score for cancer occurrence according to a support vector regression model generated from a censored and uncensored patient database, where each entry of the censored and uncensored patient database has data corresponding to the clinical, molecular and morphometric features of the patient dataset, and where the censored and uncensored patient database includes patient entries where a final health outcome for a patient is known (uncensored) and patient entries where the final health outcome for a patient is unknown (censored), and the SVRc algorithm evaluates the patient dataset according to the following:
∈( x )= W τ Φ( x )+ b;
where ∈(x) is a linear regression function on a feature space F, ∈(x) being the predicted time to event of sample x, W is a vector in F, Φ(x) is a mapping function to map x to a vector in F, b is the y-intercept,
where W and b are obtained from an epsilon-insensitive loss function:
min( W,b ) P= ½ W τ W+CΣ i=1 n (ξ i +ξ i *)
such that:
y i −( W τ Φ( x i )+ b )≦∈+ξ i
( W τ Φ( x i )+ b )− y i ≦+∈+ξ i *
ξ i ,ξ i *>0, i =1 . . . n;
where ∈, C and ξ are constants;
(6) assigning the patient to a risk of cancer occurrence based on the output value of step (5), using the processor, wherein if the output value is below 30.19, then the patient is assigned a low risk of cancer occurrence and if the output model value is above 30.19, then the patient is assigned a high risk of cancer occurrence; and
(7) generating, using the processor, a report based on the risk of cancer occurrence assigned to the patient in step (6).
3. The system of claim 1 , wherein the computer-generated morphometric measurements include one or more measurements of a minimum spanning tree (MST) identified in the one or more images of tissue subject to multiplex immunofluorescence (IF).
4. The system of claim 3 , wherein the one or more measurements of the minimum spanning tree (MST) comprises one or more measurements of a MST of epithelial nuclei identified in the one or more images of tissue subject to multiplex immunofluorescence (IF).
5. The system of claim 4 , wherein the one or more measurements of the minimum spanning tree (MST) comprises average edge length in the MST of epithelial nuclei as identified in the one or more images of tissue subject to multiplex immunofluorescence (IF).
6. The system of claim 1 , wherein the plurality of morphometric measurements include one or more measurements of the fractal dimension (FD) measured in the one or more images of tissue subject to multiplex immunofluorescence (IF).
7. The system of claim 6 , wherein the one or more measurements of the fractal dimension (FD) comprises one or more measurements of the fractal dimension of one or more glands identified in the one or more images of tissue subject to multiplex IF.
8. The system of claim 7 , wherein the one or more measurements of the fractal dimension of one or more glands comprises one or more measurements of the fractal dimension of gland boundaries between glands and stroma.
9. The system of claim 7 , wherein the one or more measurements of the fractal dimension of one or more glands comprises one or more measurements of the fractal dimension of gland boundaries between glands and stroma and between glands and lumen.