BACKGROUND
Advances in imaging techniques have lead to early detection of tumors but have had small (approximately 15-30%) impact on those malignant tumor types currently responsible for most patient mortality including lung and breast cancer. Existing techniques fail to provide quantitative and objective metrics to predict which suspect detected nodules would be found malignant if biopsied. For example, standard mammography method relies heavily on the subjective, experience, and non-quantitative judgment of highly trained mammographic radiologists. Specifically, the detection and diagnosis is based on a radiologist visually reading and interpreting two projection X-ray radiographs in the cranio-caudal (CC) and medial-lateral-oblique (MLO) orientations taken with breast compression. In addition, although some improvements have been made in existing techniques to detect smaller tumors, such improvements tend to worsen the problem of over-diagnosis, causing more harm than good. For example, by focusing on detecting the smaller tumors, more false positives (e.g., benign nodules) may also be detected, leading to more resources spent (e.g., performing additional testing or surgery) and potentially more penalties introduced (e.g., permanent loss of lung capacity due to the surgery).
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 illustrates an example method 100 of identifying a solid nodule and assessing the risks associated with the identified nodule;
FIG. 2 illustrates an example histogram of a number of solitary pulmonary nodules in the lungs of a selected population of patients that are identified with a set of mean reference Hounsfield Unit (HU) enhancement difference values;
FIG. 3 illustrates an example quantitative metric constructed from the histogram data of FIG. 2 and expressed in an annotated Receiver Operator Characteristic (ROC) curve;
FIG. 4 illustrates an example smoothing of the histogram data of FIG. 2, in accordance with one embodiment of the disclosure;
FIG. 5 illustrates an example quantitative metric constructed from the smoothed histogram data of FIG. 4 and expressed in an annotated ROC curve;
FIG. 6 illustrates an example quantitative metric constructed also from the smoothed histogram data of FIG. 4 and expressed in an annotated ROC table;
FIG. 7 illustrates an example quantitative metric constructed from the true positive fraction, false positive fraction, and specificity derived from the smoothed histogram of FIG. 4;
FIG. 8 illustrates an example histogram of a number of nodules in another anatomical site, the breasts, of a selected population of patients that are identified with a set of mean reference HU enhancement difference values;
FIG. 9 illustrates an example quantitative metric constructed from the histogram data of FIG. 8 that have been smoothed and expressed in an annotated ROC curve;
FIG. 10 illustrates an example quantitative metric constructed also from histogram data of FIG. 8 that have been smoothed and expressed in an annotated ROC table;
FIG. 11 illustrates an example quantitative metric constructed from the true positive fraction, false positive fraction, and specificity derived from the histogram data of FIG. 8 that have been smoothed;
FIG. 12 illustrates an example quantitative metric constructed from the true positive fraction, false positive fraction, and specificity derived from the smoothed histogram of FIG. 4;
FIG. 13 illustrates an example quantitative metric constructed from the positive predictive value, prevalence, accuracy, and false negative derived from the smoothed histogram of FIG. 4;
FIG. 14 illustrates an example Maximum Intensity Projection (MIP) image from a breast cancer patient;
FIG. 15 illustrates an example quantitative metric, such as the graph of FIG. 12, being used to assess multiple nodules identified in a patient being evaluated;
FIG. 16 illustrates another example quantitative metric, such as the annotated ROC curve of FIG. 9, being used to assess multiple nodules identified in a patient being evaluated;
FIG. 17 illustrates an example quantitative metric constructed from the true positive fraction, false positive fraction, and specificity derived from the histogram data of FIG. 8 and being used to assess multiple nodules identified in a patient being evaluated;
FIG. 18 illustrates an example quantitative metric from the histogram data similar to FIG. 8 and expressed in an annotated ROC curve;
FIG. 19 is a block diagram illustrating a computer program product for identifying a solid nodule and assessing the risks associated with the identified nodule; and
FIG. 20 is a schematic diagram illustrating a radiation system 2000, all arranged in accordance with some embodiments of the present disclosure.
DETAILED DESCRIPTION
The technical details set forth below enable a person skilled in the art to implement at least some embodiments of the present disclosure to identify, assess, and manage cancer diseases. In this disclosure, the term “lesion” and “nodule” are used interchangeably. Also, the term “biomarker” generally refers to a characteristic that is objectively measured and evaluated as a medical indicator of normal biologic processes, pathogenic processes, or responses to a therapeutic intervention. It should be noted a single specific biomarker's medical accuracy, precision and usefulness can be increased when used in conjunction with other biomarkers that characterize complimentary characteristics including other more biological or biochemical biomarkers such as proteins, metabolites, genomics, and others.
FIG. 1 illustrates an example method 100 of identifying a solid nodule and assessing the risks associated with the identified nodule, in accordance with one embodiment of the present disclosure. The various blocks of the method 100 are not intended to be limiting to the described embodiments. For example, one skilled in the art will appreciate that, for this and other processes and methods disclosed herein, the functions performed in the processes and methods may be implemented in differing order. Furthermore, the outlined steps and operations are only provided as examples, and some of the steps and operations may be optional, combined into fewer steps and operations, or expanded into additional steps and operations without detracting from the essence of the disclosed embodiments.
In block 102 (prepare one or more quantitative metrics for a cancer disease in a selected population of patients), relevant data for a selected population of patients for a cancer disease is collected and analyzed, so that one or more quantitative metrics for the selected population of patients may be constructed. Some examples of the relevant data may include, without limitation, numeric biomarker data associated with suspect nodules in the selected population of patients that may be acquired with or without having injected one or more biomarkers into the patients. In addition, the numeric biomarker data may be for various different anatomical sites of the patients, such as, their lungs, breasts, and others. Some example numeric biomarker data may include, without limitation, computed tomography (CT) Hounsfield Unit (HU) values for a contrast agent (e.g., iodine).
With the collected relevant data, some example quantitative metrics, such as, without limitations, sensitivity, specificity, true positive fraction (TPF), false positive fraction (FPF), Receiver Operator Characteristic (ROC) representations, positive predictive value, false negative fraction (FNF), accuracy, prevalence, and others may be constructed. Some of these quantitative metrics correspond to the following equations:
TPF (or sensitivity)=fraction of all malignancies (TP+FN) correctly diagnosed=TP/(TP+FN),
where TP corresponds to true positives, and FN corresponds to false negatives;
FPF=fraction of all benign (TN+FP) incorrectly diagnosed=FP/(TN+FP), where TN corresponds to true negatives, and FP corresponds to false positives;
Specificity=fraction of all benign (TN+FP) correctly diagnosed=TN/(TN+FP)
Positive Predictive Value=fraction of positives (TP+FP) that are true=TP/(TP+FP)
FNF=fraction of all negatives (FN+TN) that are actually malignant=FN/(FN+TN)
Accuracy=correct diagnoses (TP+TN) divided by total number of nodules=(TP+TN)/(TN+FN+TP+FP)
Prevalence=fraction of all nodules that were malignant=(TP+FN)/(TN+FN+TP+FP)
Subsequent paragraphs and figures will further detail and illustrate the construction of some of these quantitative metrics.
In block 104 (acquire contrast enhanced numeric biomarker data), according to one embodiment of the present disclosure, a first set of numeric biomarker data associated with a patient being evaluated may be acquired before the injection of the biomarker into the patient (e.g., a contrast agent such as iodine), and a second set of numeric biomarker data may be acquired after the injection of the biomarker into the patient. In an alternative embodiment, one set of numeric biomarker data may be acquired after the injection of the biomarker to reduce or eliminate the X-ray exposure dose. The biomarker's values are physically generated by the concentration gradient of the contrast medium and its physical-chemistry-diffusion through the porous membranes of the angiogenesis capillaries in cancer that changes the physical properties of the region (e.g. linear X-ray attenuation coefficients, mass densities, etc.) that are related to metrics descriptive of solid malignant tumors larger than approximately 1 mm in diameter.
In block 106 (determine HU related information from acquired numeric biomarker data), the acquired numeric biomarker data is further processed to determine HU related information (e.g., a mean HU value, a mean HU enhancement difference value, and others). The different approaches are discussed in subsequent paragraphs.
In block 108 (assess cancer risk based on HU related information and at least one of the one or more quantitative metrics), the HU related information may be utilized (e.g., the difference between the mean HU of an identified nodule with the injected biomarker and the mean HU of the same identified nodule without the injected biomarker) to help quantify cancer risk. Additional details associated with cancer risk assessment are elaborated further in subsequent paragraphs.
FIG. 2 illustrates an example histogram 200 of a number of solitary pulmonary nodules in the lungs of a selected population of patients that are identified with a set of mean reference HU enhancement difference values (ΔHUreference value). In one implementation, the vertical axis and the horizontal axis of the histogram 200 correspond to a number of nodules and a set of mean HU enhancement difference values (ΔHU values), respectively. A biopsy is performed for each of the pulmonary nodules shown in the histogram 200, indicating whether the nodule is considered to be benign or malignant. A benign nodule may be considered as a “false positive,” and a malignant nodule may be considered as a “true positive.” Also, a first set of CT data without any biomarker (e.g., iodine) enhancement and a second set of CT data with biomarker enhancement are acquired from each of the selected population of patients. For each of the pulmonary nodules shown in the histogram 200, a first mean reference HU value is calculated based on the first set of CT data, and a second mean reference HU value is calculated based on the second set of CT data. Then, the ΔHUreference value for the pulmonary nodule is obtained by subtracting the first mean reference HU value from the second mean reference HU value.
FIG. 3 illustrates an example quantitative metric constructed from the histogram data of FIG. 2 and expressed in an annotated ROC curve 300, in accordance with one embodiment of the present disclosure. The vertical axis of the annotated ROC curve 300 corresponds to TPFs, and the horizontal axis of the annotated ROC curve 300 corresponds to FPFs. TPF and FPF are discussed in earlier paragraphs. In addition, a mean threshold HU enhancement difference value (ΔHU threshold value) is placed adjacent to some of the data points plotted on the annotated ROC curve 300. Specifically, in one implementation, each ΔHUthreshold value in FIG. 3 corresponds to a certain ΔHUreference value plotted in FIG. 2, from which certain statistical relationships may be derived. For each of such annotated data points, TP corresponds to a number of true positives at or above the ΔHUthreshold value, and FP corresponds to a number of false positives also at or above the same ΔHUthreshold value.
To illustrate, suppose the ΔHUthreshold value is −6. As shown in FIG. 2, all of the FPs (i.e., benign nodules) and all of the TPs (i.e., malignant nodules) are at or above this −6 value. This corresponds to the TPF=1.0 and FPF=1.0 as shown in FIG. 3. When the ΔHUreference value goes from −6 to 20 as shown in FIG. 2, the first TP is plotted in the histogram 200. In other words, the annotated ROC curve 300 stays at TPF=1.0 until the corresponding ΔHUthreshold value reaches 20, because the fraction drops. As the ΔHUthreshold value on the annotated ROC curve 300 continues to increase, both the TPF and the FPF monotonically decrease until the ΔHUthreshold value of 60 is reached. Beyond this point, as shown in FIG. 2, there are no more remaining FPs, so the FPF in FIG. 3 then goes to zero.
FIG. 4 illustrates an example smoothing of the histogram data of FIG. 2, in accordance with one embodiment of the present disclosure. Based on the concept that medical diagnosis may be considered a random sampling process described by Normal or Gaussian probability distributions, not only is the shape of the distributions determined by the mean and standard deviation values, but a limited number of samples may also be sufficient to accurately specify means and standard deviations of randomly sampled distributions.
In one implementation, a Gaussian approximation 400 of the histogram data of FIG. 2 includes 3 distributions. Two are the benign distributions, the larger on the left with no angiogensis (i.e., the first benign distribution) and the smaller one on the right with angiogensis (i.e., the second benign distribution). The third and largest distribution where these samples are both angiogenic and malignant. All three of these Normal distributions may be multiplied by the actual number of nodules estimated to come from each distribution. This normalization makes the total number of nodules the same before and after the smoothing. The mean and standard deviation of the first benign distribution may be obtained from the data points up to the ΔHUreference value of approximately 30, and the mean and standard deviation of the second benign distribution may be obtained from the data points to the ΔHUreference value of above approximately 30. The latter may assume the same mean and standard deviation as that of the largest and malignant distribution.
FIG. 5 illustrates an example quantitative metric constructed from the smoothed histogram data of FIG. 4 and expressed in an annotated ROC curve 500, in accordance with one embodiment of the present disclosure. With the smoothed histogram data, the accuracy in the characteristics associated with the annotated ROC curve 500 may be improved.
FIG. 6 illustrates an example quantitative metric constructed also from the smoothed histogram data of FIG. 4 and expressed in an annotated ROC table 600, in accordance with one embodiment of the present disclosure.
FIG. 7 illustrates an example quantitative metric constructed from the true positive fraction, false positive fraction, and specificity derived from the smoothed histogram of FIG. 4, in accordance with one embodiment of the present disclosure. An example graph 700 includes fractions as its vertical axis and a set of ΔHU values as its horizontal axis. The resulting three curves for the selected population of patients in the graph 700 enable a person, not necessarily a highly trained radiologist, to objectively estimate the false positive, true positive, and specificity values associated with a nodule of a patient being evaluated based on the ΔHU value for the nodule and also the graph 700.
FIG. 8 illustrates an example histogram 800 of a number of nodules in another anatomical site, the breasts, of a selected population of patients that are identified with a set of ΔHUreference values.
Similar to FIG. 5, FIG. 9 illustrates an example quantitative metric constructed from the histogram data of FIG. 8 that have been smoothed and expressed in an annotated ROC curve 900, in accordance with one embodiment of the present disclosure. In one implementation, similar to the smoothing operation shown in FIG. 4 and discussed above, the histogram data of FIG. 8 are Gaussian smoothed.
Similar to FIG. 6, FIG. 10 illustrates an example quantitative metric constructed also from histogram data of FIG. 8 that have been smoothed and expressed in an annotated ROC table 1000, in accordance with one embodiment of the present disclosure.
Similar to FIG. 7, FIG. 11 illustrates an example quantitative metric constructed from the true positive fraction, false positive fraction, and specificity derived from the histogram data of FIG. 8 that have been smoothed, in accordance with one embodiment of the present disclosure.
FIG. 12 illustrates an example quantitative metric constructed from the true positive fraction, false positive fraction, and specificity derived from the smoothed histogram of FIG. 4, in accordance with one embodiment of the present disclosure. Instead of plotting against a set of ΔHU values, an example graph 1200 includes fractions as its vertical axis and a set of iodine concentration values (Ic) as its horizontal axis. In one implementation, CT HU values have been observed to be linear in iodine concentration with a 600 HU change as the iodine concentration in test samples is raised from 0 to 40 mg/mL. This corresponds to 15 HU change per mg/mL of iodine used as a conversion factor from ΔHU in Hounsfield Units to iodine concentration Ic in mg/mL.
FIG. 13 illustrates an example quantitative metric constructed from the positive predictive value, prevalence, accuracy, and false negative derived from the smoothed histogram of FIG. 4, in accordance with one embodiment of the present disclosure. Similar to FIG. 12, an example graph 1300 also includes fractions as its vertical axis and Ic as its horizontal axis.
FIG. 14 illustrates an example Maximum Intensity Projection (MIP) image 1400 from a breast cancer patient, where the largest nodule shown as Number 1 is biopsy confirmed to be malignant. The ΔHU value for each of the five nodules identified in the MIP image 1400 of a patient being evaluated may be converted to Ic values by dividing the ΔHU value by the 15 HU per iodine mg/ml conversion factor. In addition, for the values shown in FIG. 14, the higher the specificity associated with a lesion, the higher the probability that biopsy determination would show this lesion to be malignant.
FIG. 15 illustrates an example quantitative metric, such as the graph 1200 of FIG. 12, being used to assess multiple nodules identified in a patient being evaluated, in accordance with one embodiment of the disclosure. To illustrate, the five nodules identified in the MIP image 1400 are plotted in a graph 1500, and the intersections with any of the three curves (i.e., specificity, TPF, and FPF) correspond to numeric metric values.
FIG. 16 illustrates another example quantitative metric, such as the annotated ROC curve 900 of FIG. 9, being used to assess multiple nodules identified in a patient being evaluated, in accordance with one embodiment of the disclosure.
In one embodiment, one set, as opposed to two sets, of CT data set with iodine contrast is acquired, eliminating the need for the “without iodine” data set and resulting in the reduction of the X-ray exposure dose by approximately half. Rather than relying on absolute HU values, which may be difficult to calibrate, the mean HU values of adipose tissue surrounding the nodule may be used as a reference to obtain ΔHU values. FIG. 17 illustrates an example quantitative metric constructed from the true positive fraction, false positive fraction, and specificity derived from the histogram data of FIG. 8, in accordance with one embodiment of the present disclosure. An example graph 1700 includes fractions as its vertical axis and ΔHU values as its horizontal axis.
FIG. 18 illustrates an example quantitative metric from the histogram data similar to FIG. 8 and expressed in an annotated ROC curve 1800, in accordance with one embodiment of the present disclosure. Variation of adipose tissue values and the inherent variability of the glandular tissues where breast cancer nodules occur contribute to a wider range of ΔHU values (than in FIG. 8). This additional uncertainty causes the AUC of the annotated ROC curve 1800 to be lower. Suppose the first four of the five nodules identified in the MIP image 1400 are plotted in the graph 1700 as shown in FIG. 17. The uncertainty drops the ranking of the nodule 4 below that of nodule 1 in disagreement with the ranking of the same four nodules shown in FIG. 15. This change in the ranking of the nodules is consistent with the degree of confidence lost when the AUC of the annotated ROC curve 1800 drops to a lower value than the AUC of the annotated ROC curve 1600.
FIG. 19 is a block diagram illustrating a computer program product 1900 for identifying a solid nodule and assessing the risks associated with the identified nodule, in accordance with one embodiment of the present disclosure. The computer program product 1900 may include one or more sets of executable instructions 1902 for executing the methods described above and illustrated in FIG. 1. The computer program product 1900 may be transmitted in a signal bearing medium 1904 or another similar communication medium 1906. The computer program product 1900 may be recorded in a computer readable medium 1908 or another similar recordable medium 1910.
FIG. 20 is a schematic diagram illustrating a radiation system 2000, in accordance with one embodiment of the present disclosure. The radiation system 2000 includes a radiation source 2002, an electronic portal imaging device (EPID) 2006, a gantry 2010, and a control system 2016. The radiation source 2002 is aimed towards a patient 2004 and to the EPID 2006.
In the illustrated embodiment, the control system 2016 includes a processor for executing instructions, such as the executable instructions 1902 shown in FIG. 19, a monitor for displaying data, and an input device, such as a keyboard or a mouse, for inputting data. Although the control system 2016 is shown as a separate component from the gantry 2010, in alternative embodiments, the control system 2016 can be a part of the gantry 2010.
It should be noted that the radiation system 2000 should not be limited to the configuration described above, and that the system can also have other configurations.
While the forgoing is directed to embodiments of the present disclosure, other and further embodiments of the present disclosure may be devised without departing from the basic scope thereof, and the scope thereof may be determined by the claims that follow.
Patent Application
20130071334
