Systems and methods for extracting prognostic image features
Described herein are systems and methods for extracting image features from magnetic resonance imaging scans. These methods and systems are useful for both determining cancer lesion severity and disease prognosis. In particular, methods and systems are provided herein for extracting and analyzing imaging features that are correlated with breast cancer subtype and prognosis.
1. A method of treating a subject having a breast cancer disease comprising one or more breast cancer lesions, the method comprising:
a) evaluating the one or more breast cancer lesions from a magnetic resonance image (MR image) of a breast tissue from the subject, wherein the evaluation comprises:
(i) marking one or more boundaries of the one or more of the breast cancer lesions in the MR image;
(ii) correcting non-uniformity of intensity of the MR image;
(iii) segmenting the MR image into one or more tissue segments, wherein the tissue segments comprise one or more than one of a normal background parenchyma, breast cancer lesion tissue, or other breast tissue;
(iv) extracting one or more image features from the MR image, wherein the one or more extracted image features comprise one or more dynamic enhancement-based features that assess the enhancement of the breast cancer lesion tissue segment over time, the one or more dynamic enhancement-based features comprising at least one of:
(A) feature F1, which comprises a ratio of (1) an average level of breast cancer lesion image enhancement at a time point when a predetermined fraction of normal background parenchyma reaches a predetermined threshold level of image enhancement to (2) a maximum average level of breast cancer lesion image enhancement from across all time points, or
(B) feature F2, which comprises a fraction of normal background parenchyma that is above a predetermined threshold level of image enhancement at a time point when an average level of breast cancer lesion enhancement divided by a maximum average level of breast lesion enhancement from across all time points reaches a predetermined threshold; and
(v) providing a score based on the one or more extracted image features, wherein the score is indicative of a likelihood of breast cancer disease progression or recurrence, an overall survival of the subject, a recurrence-free survival of the subject, a distant recurrence-free survival of the subject, a response by the subject to neo-adjuvant therapy or a combination thereof; and
b) treating the subject based on the score, wherein treating the subject comprises initiating a standard of care medical treatment, increasing a medical treatment, decreasing a medical treatment, altering a medical treatment, or ceasing a medical treatment.
2. The method of claim 1 , wherein correcting non-uniformity of intensity comprises performing a nonparametric non-uniform intensity normalization method on the marked MR image.
3. The method of claim 1 , wherein segmentation of the normal background parenchyma is based on fat-saturated MR images, non-fat saturated MR images, or both.
4. The method of claim 1 , wherein the segment of other breast tissue is obtained by performing a global thresholding method on the MR image.
5. The method of claim 1 , wherein the normal background parenchyma and the breast cancer lesion tissue segments are obtained by performing Fuzzy C-Means clustering.
6. The method of claim 1 , wherein the one or more extracted image features further comprise one or more of:
(a) texture features;
(b) size features;
(c) shape features, either two-dimensional or three-dimensional;
(d) gradient features;
(e) features quantifying imaging-based heterogeneity of the one or more breast cancer lesions; and
(f) features quantifying spatial kinetics of the one or more breast cancer lesions.
7. The method of claim 6 , wherein the size features or shape features comprise one or more of a major axis length, a minor axis length, a mean radius length, a standard deviation of the radius length, a volume of the breast cancer lesion, a volume of the normal background parenchyma, circularity, and features based on three-dimensional analysis of the MR images.
8. The method of claim 6 , wherein the gradient features comprise a mean and a standard deviation of one or more intensity gradients of one or more breast cancer lesion boundary voxels.
9. The method of claim 6 , wherein the texture features comprise at least one or more Haralick texture features.
10. The method of claim 1 , wherein the one or more dynamic enhancement-based features comprise the feature F1.
11. The method of claim 1 , wherein the one or more dynamic enhancement-based features comprise the feature F2.
12. The method of claim 1 , wherein the feature F1 and the feature F2 are calculated according to Formula I or Formula II, respectively:
F
1
=
E
tumor
(
V
T
12
parenchyma
(
t
)
=
T
11
)
E
max
tumor
Formula
I
F
2
=
V
T
22
parenchyma
(
E
tumor
(
t
)
E
max
tumor
=
T
21
)
;
Formula
II
wherein for Formulae I and II:
E
tumor
(
t
)
=
I
tumor
(
t
)
-
I
pre
tumor
I
pre
tumor
V
T
22
parenchyma
(
t
)
=
n
(
E
parenchyma
(
t
)
>
T
22
·
E
max
tumor
)
n
parenchyma
E
parenchyma
(
t
)
=
I
parenchyma
(
t
)
-
I
pre
parenchyma
I
pre
parenchyma
;
and, wherein T 11 , T 12 , T 21 , and T 22 may each range from 0-1.
13. The method of claim 12 , wherein the one or more dynamic enhancement-based features comprise the feature F1 and wherein T 11 is 0.05 and T 12 is 0.5 or T 11 is 0.05 and T 12 is 0.1.
14. The method of claim 12 , wherein the one or more dynamic enhancement-based features comprise the feature F2 and wherein T 21 is 0.8 and T 22 is 0.5 or T 21 is 0.5 and T 22 is 0.1.
15. The method of claim 12 , wherein the one or more dynamic enhancement-based features comprise the feature F1 and wherein a value of the feature F1 greater than 0.75 is indicative of an increased risk of breast cancer disease recurrence.
16. The method of claim 12 , wherein the one or more dynamic enhancement-based features comprise the feature F1 and wherein:
a) a value of the feature F1 greater than 0.75; and
b) a peak enhancement location before a second post-contrast sequence,
together are indicative of luminal A or luminal B breast cancer.
17. The method of claim 12 , wherein the one or more dynamic enhancement-based features comprise the feature F1 and wherein:
T 11 is 0.05 and T 12 is 0.1; or
T 11 is 0.02 and T 12 is 0.5; or
T 11 is 0.05 and T 12 is 0.5; or
T 11 is 0.1 and T 12 is 0.5; or
T 11 is 0.05 and T 12 is 0.8; or
T 11 is 0.1 and T 12 is 0.8.
18. The method of claim 1 , wherein the medical treatment comprises chemotherapy, hormonal therapy, immunotherapy or combinations thereof.
19. The method of claim 1 , wherein the score, alone or in combination with genomic and pathological profile, is used to determine if patients will receive chemotherapy.
20. The method of claim 1 , further comprising combining the score with one or more of cancer lesion receptor status, cancer lesion receptor status-based lesion subtype, lesion size, lymph node involvement, commercial genomic assays, PAM50 marker, lesion pathology, other tumor markers, or genomic markers.
21. The method of claim 1 , wherein the subject has a breast cancer that is a lymph-node positive or lymph-node negative breast cancer.
22. The method of claim 1 , wherein the breast cancer lesion is ER-positive or -negative, PR-positive or -negative, HER2-positive or -negative, or any combination thereof, or wherein the breast cancer lesion is luminal A, luminal B, basal, normal-like or claudin-low.
23. The method of claim 17 , wherein the breast cancer lesion is large, ER-positive and/or PR-positive, and HER2-negative.
24. The method of claim 23 , wherein a value of the feature F1 greater than or equal to 0.75 is indicative of a lower likelihood of recurrence-free survival.