Aleksandar Danicic

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of United States Provisional

Application Serial No. 62/326,006 by Aleksandar Danicic, filed April 22, 2016 and entitled "Methods for Thermal Breast Cancer Detection," the contents of which are hereby incorporated by reference in their entirety in this application.

FIELD OF THE INVENTION

[0002] The invention relates generally to thermographic breast cancer detection methods, and more particularly to breast preparation and digital processing of thermographic image data for the detection of breast cancer. The invention is specifically aimed at providing a simple and cost-effective method for high-probability non-invasive breast screening and early detection.

BACKGROUND OF THE INVENTION

[0003] Breast cancer is the leading type of cancer in women worldwide, accounting for approximately twenty-five percent of all recorded cases [1 ]. In the past several decades, a number of breast cancer screening techniques have been developed including: clinical- and self-examination, mammography, ultrasound, magnetic resonance imaging, genetic screening, and thermography. [0004] Breast examination is not considered reliable in screening of women without symptoms and at low risk for developing breast cancer as it results in

substantial positive false alarms (e.g. due to benign lesions) and unnecessary biopsies [2]. Mammography is the most common screening method today and widely used in developed countries [3]. Mammography is found to be marginally useful in detecting breast tumors in women under forty years of age due to dense breast tissue

characteristics of younger women [4,5]. Yet, most aggressive breast tumors are found to thrive in dense breast tissue. Ultrasound is often employed as a diagnostic aid to mammography for dense tissue breast cancer detection, though it suffers from low resolution and increased false alarm rate [6,7]. Magnetic resonance imaging (MRI) can be used to detect tumors invisible to mammograms and generally has a high rate of negative predictive value (i.e. ruling out presence of cancer with high degree of certainty). On the other hand, MRI breast cancer screening still suffers from a high rate of false alarms [8]. In addition, MRI's are very expensive procedures and cannot be performed on all women (e.g., patients with pacemakers, tissue expanders, and other modifications affecting the chest area or breast tissue). Lastly, genetic testing does not detect breast tumors directly, but may reveal a tendency toward cancer development by detecting genetic markers that are associated with the development of breast cancer [9].

[0005] Thermography is a screening method based on interpreting

thermograms-infrared images generated using a camera which captures infrared thermal radiation emitted by human body in the 8-14 micron wavelength range.

Thermographs provide heat surface maps of breast(s), pinpointing areas with elevated temperatures. Increased blood vessel or chemical activity is typically present in precancerous and cancerous tissue as tumors require a larger nutrient supply than healthy cells; the nutrient supply is delivered via increased circulation through existing blood vessels, open dormant vessels, and newly created blood vessels [10]. [0006] Thermography has recently seen increasing interest in breast cancer screening due to its low cost and non-invasive nature. Medical practitioners and researchers will typically use expensive large thermal imaging arrays due to their high spatial resolution [1 1 ]. Recently, low-cost lower-resolution thermal imaging cameras have appeared on the consumer market, primarily targeting non-medical applications. A large amount of controversy and disagreement still surrounds the utility of

thermograms in breast cancer screening. A high rate of false positive detections (i.e. low specificity) and inability to discern smaller (less than 1 cm in diameter) and/or buried tumors plague this otherwise very attractive screening [12, 13, 14]. Thus, there exists a need for substantial increase in the fidelity of thermal breast cancer screening and early detection.

SUMMARY OF THE INVENTION

[0007] The present invention advantageously fills the aforementioned

deficiencies by uniquely providing a method for high-fidelity thermal breast cancer detection. Specifically, the invention outlines a series of methods which can be applied independently, or in conjunction with one another, in order to: (i) reduce the probability of false positive detections, and (ii) detect increasingly smaller and more deeply buried tumors.

[0008] Embodiments include a method for transient thermal analysis which can be used for high accuracy initial screening for potential breast tumors. Inverse heat transfer computation techniques may be used to further improve ability to detect increasingly smaller and deeply buried tumors with high sensitivity and specificity.

Breast health monitoring is accomplished by periodic thermogram capture and applying image processing techniques described herein. Individual images as well as their time series may be processed and provide means of early breast cancer risk detection.

[0009] One aspect of the invention is a method of performing thermal breast cancer detection comprising the steps of: (1 ) providing a patient to undergo thermal breast cancer detection;

(2) cooling the surface of a breast of the patient;

(3) capturing one or more thermal images from one or more angles from the breast of the patient;

(4) performing a segment-by-segment analysis of the thermal images employing inverse heat transfer analysis to calculate the probability of positive breast cancer identification;

(5) determining whether the probability of positive cancer identification calculated in step (4) exceeds a predefined threshold; and

(6) if the determination of positive cancer identification calculated in step (5) is equal to or exceeds the predefined threshold, notifying the patient of the result.

[0010] In one alternative, the threshold is selected from the group consisting of 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, 95%, and 98%. In another alternative, the threshold is varied depending on at least one variable selected from the group of: age of the patient, weight of the patient, history of alcohol consumption by the patient, history of tobacco use by the patient, the existence of a mutation in a known breast cancer marker gene such as BRCA 1 or BRCA2 if the existence of the mutation is known for the patient, and the ethnic group identification of the patient if the existence of the mutation is not known for the patient.

[0011] Another aspect of the invention is a method of performing thermal breast cancer detection comprising the steps of:

(1 ) providing a patient to undergo thermal breast cancer detection;

(2) cooling the surface of a breast of the patient;

(3) capturing one or more thermal images from one or more angles from the breast of the patient;

(4) storing the thermal image data from step (3) in a storage device;

(5) transferring the stored thermal image data from step (3) to a thermogram database; (6) performing digital breast registration and segmentation on the stored thermal image data from step (4);

(7) performing inverse heat transfer analysis on the results of digital breast registration and segmentation from step (6);

(8) performing temporal statistical analysis on the results of inverse heat transfer analysis from step (7);

(9) applying statistical classification techniques to the results of temporal statistical analysis from step (8) with the inclusion of the data in the thermogram database from step (5);

(10) calculating the probability of positive cancer identification from the results of applying statistical classification from step (9);

(1 1 ) determining whether the probability of positive cancer identification calculated in step (10) exceeds a predefined threshold;

(12) if the probability of positive cancer identification calculated in step (10) is equal to or exceeds the predefined threshold as determined in step (1 1 ), notifying the patient of the result;

(13) if the probability of positive cancer identification calculated in step (10) is less than the predefined threshold as determined in step (1 1 ), repeating steps (2)-(1 1 ) by performing an additional thermogram until at least 10 thermograms are performed.

[0012] Suitable thresholds can be determined as described above.

[0013] Preferably, the cooling step (2) is performed without the presence of moisture. Preferably, the cooling step (2) is performed by a method selected from the group consisting of applying a cool fabric to the breast surface and using a fan to cool the breast surface.

[0014] The method can further comprise the steps of: (i) performing transient heating of the breast; and (ii) capturing one or more thermal images subsequent to the performance of transient heating of the breast. [0015] In one alternative, the thermal images are captured from a single angle. However, a generally preferred alternative is for the thermal images to be captured from multiple angles. When the thermal images are captured from multiple angles, the thermal images captured from multiple angles can be used to reconstruct three- dimensional breast surface thermal profiles of high accuracy. In another alternative, the use of thermal images captured from multiple angles is coupled with attitude and heading reference systems that are based on one or more sensors. Typically, the sensor is located in a device selected from the group consisting of a gyroscope, an accelerometer, and a magnetometer.

[0016] Typically, the inverse heat transfer computation technique is selected from the group consisting of the regularized conjugate gradient, Bayesian, adjoint, and hybrid methods.

[0017] Typically, multiple thermograms are captured for the same patient periodically over time. The interval over which the multiple thermograms are taken can be, but is not limited to, an interval of from 1 to 60 minutes, 1 .5 to 24 hours, or 1 .5 to 30 days. The taking of multiple thermograms on a single patient can produce a time series and a storable database. The database can be stored locally, or can be stored centrally. The database can be used to track and update temporal statistics of images.

[0018] Temperature variations of each breast segment can be tracked based on an adaptive filter theory principle. Typically, the adaptive filter theory principle is selected from the group consisting of least mean squares (LMS) and recursive least squares (RLS).

[0019] In one alternative, the centrally stored database is used as a dataset from which a cancer detection algorithm is further trained to more accurately classify breast tumors. The cancer detection algorithm is typically selected from the group consisting of machine learning and pattern recognition techniques. The cancer detection algorithm is typically based on a feature selected from the group consisting of: (i) relative temperature differences between two breasts; (ii) relative difference and cross- correlation between neighboring segments in the same breast; (iii) a statistical parameter selected from the group consisting of mean, variance, skewness, and kurtosis; and (iv) entropy of breast segments. The feature can be used to train a machine learning algorithm that is selected from the group consisting of logistic regression, support vector machines, and neural networks.

[0020] Typically, the analysis of the thermograms is performed with a

discriminator value (deltaT) selected from the group consisting of 0.025-, 0.050-, and 0.075-degree Celsius. In another alternative, the analysis of the thermograms is performed with a discriminator value (deltaT) that is selected based on a pre-test risk of breast cancer in the patient to be tested, depending on one or more factors selected from the group consisting of age, weight, history of alcohol consumption, history of tobacco use, mutation status for BRCA 1 or BRCA2, and ethnic group in patients for which the status of mutations in BRCA 1 or BRCA2 is not known.

[0021] The methods described above can further comprise the step of detecting at least one biomarker associated with breast cancer. The at least one biomarker can be selected from the group consisting of uPA, PAI-1 , TF, thioredoxin, a gene product associated with miR-21 or miR-17-5p, TOX3 protein, cytosolic serine hydroxymethyl transferase (cSHMT), utrophin, human inter alpha trypsin inhibitor heavy chain H4 (ITIH4) fragment 1 b (BC-1 b), ER/PR (estrogen receptor / progesterone receptor), estrogen-related receptor alpha, mucin 1 , carcinoembryonic antigen, c-erbB-2, and HER2 (human epidermal growth factor 2).

BRIEF DESCRIPTION OF THE DRAWINGS [0022] These and other features, aspects, and advantages of the present invention will become better understood with reference to the following description, appended claims, and accompanying drawings where:

[0023] Figure 1 shows a generic layout of the method for thermal breast cancer detection.

[0024] Figure 2A shows a thermal image of breast taken without surface cooling. The arrow points to a tumor positively identified by biopsy.

[0025] Figure 2B shows a thermal image of breast shown in Figure 2A with surface cooling applied prior to image acquisition. The greater amount of detail in image clearly reveals a local tumorous hot spot as the surface is warmed by internal heat source due to increased vascular activity.

[0026] Figures 3A-C show a breast thermogram of a female subject taken at the beginning, middle, and end of a two-month study period. Figure 3A was taken at the beginning of the period; Figure 3B was taken at the middle of the period; and Figure 3C was taken at the end of the period.

[0027] Figure 4A shows a difference image between two thermograms prior to image registration.

[0028] Figure 4B shows a difference image between two thermograms following successful automated image registration.

[0029] Figure 5 shows a probability distribution function of the time series of the temperature difference between a single pixel on the breast and the average

temperature of its nearest neighbors. [0030] Figure 6 shows the probability of true positive detection (on the y-axis) as a function of the number of consecutive thermograms. Thermograms were taken every other day and a thermal hotspot was introduced adding deltaT temperature to a single pixel in order to mimic an appearance of hypervascular cancerous activity. The depicted probability was derived simply using a priori knowledge of the Gaussian fitted probability distribution function of healthy breast thermogram time series as shown in Figure 5 and assuming that incoming samples with introduced hotspot simply obey the same Gaussian probability distribution function with an upshifted mean.

DETAILED DESCRIPTION OF THE INVENTION

[0031] The quality of any diagnostic method, such as a method that attempts to determine the presence or absence of breast cancer, is determined by the proportion of false negative and false positive results obtained. The ideal, which is not attainable in practice, is to have a false negative rate and a false positive rate of zero; in that case, every patient screened in which the method indicates the presence of breast cancer actually has the disease, while every patient screened in which the method indicates the absence of breast cancer is free of the disease when the test was performed. A false negative result is a test result in which a patient screened by the test and in which the test indicates that the patient is free of the disease actually has the disease, breast cancer in this instance. A false positive result is a test result in which a patient screened by the test and in which the test indicates that the patient has the disease actually is free of the disease. The sensitivity of the test is the proportion of patients who test positive for the disease among those who actually have the disease; the higher the sensitivity, the lower the proportion of false negative results. The specificity of the test is the proportion of patients who test negative for the disease among those who actually are free of the disease; the higher the specificity, the lower the proportion of false positive results. The consequences of false positive and false negative results with respect to breast cancer are significantly different. A false positive result, although it can have temporary psychological consequences for the patient receiving the false positive result, can be corrected by rescreening by another test method, and, once the absence of breast cancer is confirmed by additional tests, there are no significant long- term consequences. A false negative result in a patient in which breast cancer exists, on the other hand, can have more serious consequences, as an undiagnosed

malignancy will continue to grow without treatment and may be untreatable or may require treatment with a lower probability of success or more serious side effects by the time the malignancy is actually diagnosed. Therefore, it is the goal of any diagnostic test to reduce false negatives as much as is practicable, while keeping the number of false positives to an acceptable level.

[0032] In one embodiment of the present invention, the breast surface is cooled prior to capturing one or a series of thermal images. The cooling of breast surface more accurately reveals internal heat sources as the image is captured during transient rather than steady-state heat transfer from the potential cancerous regions to the surface. Ideally, the applied cooling is free of moisture, as liquid droplets tend to cause smearing of thermal image due to scattering and varying surface emissivity profile. Breast surface cooling can be accomplished by applying a cool dry towel to the breast surface, using a fan to cool the breast surface, or any other method which will allow for reduction of surface temperature of the breasts.

[0033] In another embodiment, a series of thermal images is captured during the transient heating of the breasts (i.e. following the aforementioned cooling procedure). This will further improve the specificity of breast cancer detection by analyzing the difference in heating rates between regions of high vascular activity, which may be associated with malignancy, and normal vascular activity.

[0034] In yet another embodiment, thermograms are captured from one or multiple angles. Image capture from multiple angles provides for improved spatial resolution of curved breast surfaces. Furthermore, multi-angle thermograms can be used to reconstruct three-dimensional breast surface thermal profiles of high accuracy, especially when coupled with attitude and heading reference systems based on one or more sensors such as, but not limited to, a gyroscope, an accelerometer, and/or a magnetometer. Such sensors are commonly found in mobile platforms, e.g. mobile phones and tablets.

[0035] In yet another embodiment, inverse heat transfer computation techniques are used to derive temperature distribution internal to the breast from the captured surface temperature profile. This technique can be particularly useful in identifying small and/or deeply buried tumors. A reconstructed three-dimensional temperature surface profile is used as a starting point for application of inverse thermal conduction methods. Inverse problems are ill-posed and require regularization [15]. Application of an inverse method requires a starting thermo-physical model of the breast which models the fatty, glandular, skin, and potential cancerous tissue. The starting thermo- physical model is an approximation which is further refined during the numerical computation process. The common inverse problem techniques that could be used are the regularized conjugate gradient, Bayesian, adjoint, and/or hybrid methods. Other methods could be easily identified and utilized by those skilled in the art.

[0036] In yet another embodiment, thermograms are taken periodically over time, building up a database which is stored either locally (e.g. mobile phone, tablet, and/or computer) or centrally (e.g. cloud or dedicated server). The thermogram database is used to track and update temporal statistics of images, which in itself increases the predictive accuracy as compared to images taken at a single time instance. For example, temperature level variation between each breast segment and the average of its nearest neighbors is computed for each image in the time series. A probability distribution function is computed and used to identify and classify newly captured samples as either normal expected values or potential cancerous anomalies. Probabilities of true positive, true negative, false positive, and false negative detection are calculated with each incoming thermogram. An alarm threshold can therefore be set for when a user should be notified of a high probability of potential anomaly requiring further medical attention. For example, the alarm threshold can be 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, 95%, or 98%. The alarm threshold can also be set at an intermediate value. The alarm threshold can also be varied depending on variables such as the age of the user, weight of the user, history of alcohol consumption by the user, history of tobacco use by the user, the existence of a mutation in a known breast cancer marker gene such as BRCA1 or BRCA2 if the existence of the mutation is known for the user, or the ethnic group identification of the user if the existence of the mutation is not known for the user, as the frequency of mutations in BRCA 1 or BRCA2 varies among ethnic groups.

[0037] In order to illustrate the effectives of the aforementioned method, the inventor has performed a study on a twenty-six-year-old female subject. The subject was instructed to take thermograms using a low-cost consumer-grade thermal camera (FLIR ONE) once every morning and evening for a two-month period. A small subset of captured thermal images is shown in Figure 3. The thermal images were normalized to have the same dynamic range, spanning 76.8 to 100.0 degree Fahrenheit. Due to varying times and dates at which the images were taken, we note the stark difference in temperature distribution and average temperatures of the breasts. The next step in processing of the thermograms was image registration, which is needed to perfectly overlap the two images prior to further analysis. Figure 4A depicts a difference image between two raw thermograms and Figure 4B shows the difference image following successful automated image registration. In a similar manner, another eighty images were collected and registered against the first captured image, producing a spatially- aligned time series of thermograms. Next, a region in the middle of the right breast was selected to perform a time series analysis of the temperature difference between a single pixel (corresponding to an approximately 3 mm region) and its nearest neighbors. The calculated histogram is shown in Figure 5, fitted with a Gaussian (i.e. normal) distribution function. Despite the fact that the average temperature difference among images is several degrees Celsius, it is noted that the time series of temperature difference of a single pixel and its nearest neighbors exhibits a standard deviation of only 0.1 degrees Celsius. Finally, a temperature increase ("hotspot") was artificially added to subsequent thermograms to mimic an appearance of a minute breast tumor. By employing knowledge of the previously calculated statistical behavior of this particular region of the breast, the probability of accurate positive detection, which is inversely proportional to the probability of false alarm, can be calculated. This probability is shown in Figure 6 for 0.025-, 0.050-, and 0.075-degree Celsius added hotspots. As expected, the probability increases with each incoming thermogram, reaching ninety percent after the tenth thermogram for a hotspot as small as 0.025- degree Celsius. The significance of this result cannot be overstated, as even the noise of the thermal camera itself specified by the manufacturer is two times larger (0.050- degree). It is possible, of course, to use thermal cameras with lower noise.

[0038] In yet another embodiment, temperature variations of each breast segment are tracked based on adaptive filter theory principles such as least mean squares (LMS) or recursive least squares (RLS) [16]. LMS is an adaptive linear filter based on a stochastic gradient descent method. In particular, LMS is adapted using the current prediction error, where the error is defined as the difference between the current and the predicted temperature levels. The input signals to the algorithm are considered to be stochastic. RLS is an adaptive linear filter which recursively computes its filter coefficients based on minimization of a weighted linear least squares cost function. In contrast to LMS, the input signals are considered to be deterministic where the algorithm aims to reduce the mean square error. In the context of temporal analysis of thermograms, the LMS-based (or RLS-based) algorithm adaptively tracks variations in the first order (the mean value) and the second order (the autocorrelation values) statistics of the underlying random process which models temperature variations for each breast segment.

[0039] In yet another embodiment, the thermogram database is used as a large, and constantly growing, dataset from which cancer detection algorithms are further trained to more accurately classify breast tumors. For example, classification

algorithms utilizing machine learning and pattern recognition techniques are two clear choices for exploiting the growing dataset. In the context of tumor detection and classification, various image features can be utilized in training of supervised learning algorithms. The proposed features are: (i) relative temperature differences between two breasts; (ii) relative difference and cross-correlation between neighboring segments in the same breast; (iii) statistical parameters such as mean, variance, skewness, and kurtosis; and (iv) entropy of breast segments. The generated features are subsequently used to train well-known machine learning algorithms such as logistic regression, support vector machines, and (deep) neural networks.

[0040] Figure 1 is a flowchart showing a generic layout of the process for thermal breast cancer detection. In Figure 1 , the process begins with the start of thermal breast cancer screening 100. The breast surface is then cooled 101 . Thermal images are then captured from one or more angles 102. The thermal image data is then stored in a device capable of digital storage 103 to produce a thermogram database 104. Digital breast registration and segmentation 105 is then performed on the stored thermal image data 103. The digital data that has been subjected to breast registration and segmentation 105 is then subject to an inverse heat transfer analysis 106. Temporal statistical analysis 107 is then performed. The data in the thermogram database 104 is then combined with the data subjected to temporal statistical analysis 107 and statistical classification techniques are applied 108; this is a recursive process. The probability of a positive cancer identification is then determined 109. The next step is to determine whether the probability of a positive cancer identification 109 exceeds a set threshold 1 10. If the probability of positive cancer identification 109 equals or exceeds the set threshold 1 10, the user is alerted 1 1 1 . If the probability of positive cancer identification 109 is less than the set threshold 1 10, the user captures the next thermogram 1 12; the results from the next thermogram 1 12 are stored as thermal image data 103, and the analytical steps 105, 106, 107, and 108 are repeated.

[0041] Figure 2A shows a thermal image of breast taken without surface cooling. The arrow points to a tumor positively identified by biopsy.

[0042] Figure 2B shows a thermal image of breast shown in Figure 2A with surface cooling applied prior to image acquisition. Greater amount of detail in image clearly reveals a local tumorous hot spot as the surface is warmed by internal heat source due to increased vascular activity.

[0043] Figures 3A-C show a breast thermogram of a female subject taken at the beginning, middle, and end of a two-month study period. Figure 3A was taken at the beginning of the period; Figure 3B was taken at the middle of the period; and Figure 3C was taken at the end of the period.

[0044] Figure 4A shows a difference image between two thermograms prior to image registration.

[0045] Figure 4B shows a difference image between two thermograms following successful automated image registration.

[0046] Figure 5 shows a probability distribution function of the time series of the temperature difference between a single pixel on the breast and the average

temperature of its nearest neighbors.

[0047] Figure 6 shows the probability of true positive detection (on the y-axis) as a function of the number of consecutive thermograms. Thermograms were taken every other day and a thermal hotspot was introduced adding deltaT temperature to a single pixel in order to mimic an appearance of hypervascular cancerous activity. The depicted probability was derived simply using a priori knowledge of the Gaussian fitted probability distribution function of healthy breast thermogram time series as shown in Figure 5 and assuming that incoming samples with introduced hotspot simply obey the same Gaussian probability distribution function with an upshifted mean.

[0048] One aspect of the invention is a method of performing thermal breast cancer detection comprising the steps of:

(1 ) providing a patient to undergo thermal breast cancer detection;

(2) cooling the surface of a breast of the patient; (3) capturing one or more thermal images from one or more angles from the breast of the patient;

(4) performing a segment-by-segment analysis of the thermal images employing inverse heat transfer analysis to calculate the probability of positive breast cancer identification;

(5) determining whether the probability of positive cancer identification calculated in step (4) exceeds a predefined threshold; and

(6) if the determination of positive cancer identification calculated in step (5) is equal to or exceeds the predefined threshold, notifying the patient of the result.

[0049] As detailed above, the threshold can be 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, 95%, or 98%. The alarm threshold can also be set at an intermediate value. The alarm threshold can also be varied depending on variables such as the age of the user, the existence of a mutation in a known breast cancer marker gene such as BRCA 1 or BRCA2 if the existence of the mutation is known for the user, or the ethnic group identification of the user if the existence of the mutation is not known for the user, as the frequency of mutations in BRCA 1 or BRCA2 varies among ethnic groups.

[0050] Preferably, this aspect of the present invention is a method of performing thermal breast cancer detection comprising the steps of:

(1 ) providing a patient to undergo thermal breast cancer detection;

(2) cooling the surface of a breast of the patient;

(3) capturing one or more thermal images from one or more angles from the breast of the patient;

(4) storing the thermal image data from step (3) in a storage device;

(5) transferring the stored thermal image date from step (4) to a thermogram database;

(6) performing digital breast registration and segmentation on the stored thermal image data from step (4); (7) performing inverse heat transfer analysis on the results of digital breast registration and segmentation from step (6);

(8) performing temporal statistical analysis on the results of inverse heat transfer analysis from step (7);

(9) applying statistical classification techniques to the results of temporal statistical analysis from step (8) with the inclusion of the data in the

thermogram database from step (5);

(10) calculating the probability of positive cancer identification from the results of applying statistical classification from step (9);

(1 1 ) determining whether the probability of positive cancer identification calculated in step (10) exceeds a predefined threshold;

(12) if the probability of positive cancer identification calculated in step (10) is equal to or exceeds the predefined threshold as determined in step (1 1 ), notifying the patient of the result;

(13) if the probability of positive cancer identification calculated in step (10) is less than the predefined threshold as determined in step (1 1 ), repeating steps (2)-(1 1 ) by performing an additional thermogram until at least 10 thermograms are performed.

[0051] As detailed above, the analysis of the thermograms can be performed with a discriminator value (deltaT) that is 0.025-, 0.050-, or 0.075-degree Celsius. Other discriminator values can also be used. The lower the discriminator value, the lower the proportion of false negatives, at the risk of increasing the proportion of false positives. The higher the discriminator value, the higher the proportion of false negatives, while reducing the proportion of false positives. The discriminator value can be selected based on a pre-test risk of breast cancer in the patient to be tested, depending on one or more factors selected from the group consisting of age, weight, history of alcohol consumption, history of tobacco use, mutation status for BRCA 1 or BRCA2, and ethnic group in patients for which the status of mutations in BRCA1 or BRCA2 is not known. [0052] Preferably, the step of digital breast registration and segmentation in step (6) is performed using automated registration to ensure precise alignment in the course of registration.

[0053] In this alternative of the method, the breast surface is cooled prior to capturing one or a series of thermal images. The cooling of the breast surface more accurately reveals internal heat sources as the image is captured during transient rather than steady-state heat transfer from the potential cancerous regions to the surface; this improves the resolution of the technique. Preferably, the applied cooling is free of moisture, as the presence of moisture such as liquid droplets tends to cause smearing of the thermal image due to scattering and variations in the surface emissivity profile. Cooling of the breast surface can be accomplished by, for example, applying a cool dry towel or other fabric to the breast surface, using a fan to cool the breast surface, or by using another method that accomplishes reduction of the surface temperature of the breasts that are examined.

[0054] In another alternative, the method can further include steps of: (i) performing transient heating of the breast; and (ii) capturing one or more thermal images subsequent to the performance of transient heating of the breast. This alternative can further improve the specificity of breast cancer detection by analyzing the difference in heating rates between regions of high vascular activity and normal vascular activity.

[0055] The thermograms (thermal images) can be captured from a single angle or from multiple angles. Image capture from multiple angles provides for increased spatial resolution of curved breast tissue. In another alternative, multi-angle

thermograms are used to reconstruct three-dimensional breast surface thermal profiles of high accuracy. Typically, the use of multiple-angle thermograms is coupled with attitude and heading reference systems that are based on one or more sensors. The sensor can be, but is not limited to, a gyroscope, an accelerometer, or a magnetometer. Such sensors are typically found in mobile platforms such as a mobile phone or a tablet. [0056] As stated above, the method typically employs inverse heat transfer computation techniques to derive a temperature distribution internal to the breast from the captured surface temperature profile of the breast. The use of such inverse heat transfer computation techniques can be particularly useful in identifying small or buried tumors. In this alternative, a reconstructed three-dimensional temperature surface profile is used as a starting point for application of inverse thermal conduction methods. Inverse heat transfer problems have been studied, and generally involve estimation of a surface heat flux outgoing from temperature data measured inside a three-dimensional body which can be rigid or plastic. In general, the application of an inverse method requires a starting thermo-physical model of the breast which models the fatty, glandular, skin, and potential cancerous tissue. The starting thermo-physical model is an approximation which is further refined during the numerical computation process. The common inverse problem techniques that could be used, include, but are not limited to, a technique selected from the group consisting of the regularized conjugate gradient, Bayesian, adjoint, and hybrid methods. Other methods could be easily identified and utilized by those skilled in the art.

[0057] In methods according to the present invention, a single thermogram can be taken at a single point in time. However, it is generally preferred, and leads to increased sensitivity, to take multiple thermograms for the same patient periodically over time. The interval over which the multiple thermograms can be taken can be, but is not limited to, 1 to 60 minutes, 1 .5 to 24 hours, 1 .5 to 30 days, or longer. The taking of multiple thermograms on a single patient produces a time series. This also builds up a database that can be stored. In one alternative, the database is stored locally, such as in a mobile phone, a personal digital assistant, a tablet, or a laptop or desktop computer. In another alternative, the database is stored centrally, such as in a cloud or in a dedicated server; results from multiple patients can be stored centrally. The cloud or dedicated server can communicate with other devices through conventional communication channels. The thermogram database is used to track and update temporal statistics of images, which in itself increases the predictive accuracy as compared to images taken at a single time instance. For example, temperature level variation between each breast segment and the average of its nearest neighbors is computed for each image in the time series. A probability distribution function is computed and used to identify and classify newly captured samples as either normal expected values or potential cancerous anomalies.

[0058] In another alternative, temperature variations of each breast segment are tracked based on an adaptive filter theory principle. The adaptive filter theory principle employed for tracking and analysis of the temperature variations of each breast segment can be, but is not limited to, least mean squares (LMS) or recursive least squares (RLS). LMS is an adaptive linear filter based on a stochastic gradient descent method. In particular, LMS is adapted using the current prediction error, where the error is defined as the difference between the current and the predicted temperature levels. The input signals to the algorithm are considered to be stochastic. RLS is an adaptive linear filter which recursively computes its filter coefficients based on minimization of a weighted linear least squares cost function. In contrast to LMS, the input signals are considered to be deterministic where the algorithm aims to reduce the mean square error, !n the context of temporal analysis of thermograms, the LMS-based (or RLS- based) algorithm adoptively tracks variations in the first order (the mean value) and the second order (the autocorrelation values) statistics of the underlying random process which models temperature variations for each breast segment.

[0059] In yet another alternative, when the thermogram database is stored centrally, the thermogram database is used as a large, and constantly growing, dataset from which cancer detection algorithms are further trained to more accurately classify breast tumors. For example, classification algorithms utilizing machine learning and pattern recognition techniques are two clear choices for exploiting the growing dataset. In the context of tumor detection and classification, various image features can be utilized in training of supervised learning algorithms. The proposed features are: (i) relative temperature differences between two breasts, (ii) relative difference and cross- correlation between neighboring segments in the same breast, (iii) statistical parameters such as mean, variance, skewness, and kurtosis, and (iv) entropy of breast segments. The generated features are subsequently used to train well-known machine learning algorithms such as logistic regression, support vector machines, and (deep) neural networks.

[0060] Any of the aforementioned embodiments can be used individually or in any combination to provide an increase in the fidelity of thermal breast cancer detection.

[0061] In another aspect of the invention, the thermal detection method as described above can be used together with the detection of one or more biomarkers associated with breast cancer in order to improve the accuracy of diagnosis.

Biomarkers associated with breast cancer and that can be used for diagnosis include uPA (a serine protease), PAI-1 (an inhibitor of uPA), and TF (an aberrantly glycosylated carbohydrate and cancer-associated antigen). Other biomarkers associated with breast cancer and that can be used for diagnosis include thioredoxin, a gene product associated with miR-21 or miR-17-5p, TOX3 protein, cytosolic serine hydroxymethyl transferase (cSHMT), utrophin, human inter alpha trypsin inhibitor heavy chain H4 (ITIH4) fragment 1 b (BC-1 b), ER/PR (estrogen receptor / progesterone receptor), estrogen-related receptor alpha, mucin 1 , carcinoembryonic antigen, c-erbB-2, and HER2 (human epidermal growth factor 2). Detection and quantitation of these biomarkers is to be used together with the thermal detection method to improve diagnostic accuracy. Methods for detection of these protein biomarkers are well known in the art and include immunoassays such as include radioimmunoassay, ELISA

(enzyme-linked immunosorbent assay), competitive immunoassay, immunoassay employing lateral flow test strips, western blot assay following gel electrophoresis and other assay methods.

[0062] In one alternative, the images are uploaded directly to the server of a medical provider, such as a server located in a doctor's office, hospital, or clinic, and processed by that server. In another alternative, the images are stored and processed directly on a user's smartphone, tablet, laptop computer, or desktop computer without further uploading; only the result is transmitted to the server of the medical provider. The connection between the user's smartphone, tablet, laptop computer, or desktop computer is secure so that unauthorized individuals cannot access the information.

[0063] The following references are cited in this patent application by number. These references are incorporated herein by this reference. These references are not necessarily prior art:

[1 ] "World cancer report 2014," World Health Organization, 2014.

[2] J. P. Kosters, P. C. G0tzsche, "Regular self-examination or clinical examination for early detection of breast cancer," Cochrane Database Syst Rev, 2003.

[3] P. C. G0tzsche, K. J. J0rgensen, "Screening for breast cancer with mammography", Cochrane Database Syst Rev, 2013.

[4] "The Mayo clinic breast cancer book". Rosetta Books, pp. 124, 2012.

[5] H. Reynolds, "The Big Squeeze: a social and political history of the controversial mammogram," Cornell University Press, pp. 77, 2012.

[6] W. A. Berg, J. D. Blume, J. B. Cormack, et al, "Combined screening with ultrasound and mammography vs mammography alone in women at elevated risk of breast cancer", Journal of American Medical Association, 2008.

[7] W. A. Berg, Z. Zhang, D. Lehrer, et al, "Detection of breast cancer with addition of annual screening ultrasound or a single screening MRI to mammography in women with elevated breast cancer risk," Journal of American Medical Association, 2012.

[8] J. Hrung, S. Sonnad, J. Schwartz, and C. Langlotz, "Accuracy of MR imaging in the work-up of suspicious breast lesions: a diagnostic meta-analysis," Academic Radiology, 1999.

[9] "Genetic Risk Assessment and BRCA Mutation Testing for Breast and Ovarian Cancer Susceptibility: Recommendation Statement," Agency for Healthcare Research and Quality (United States Preventive Services Task Force), 2005.

[10] H. Qi, P. T. Kuruganti, "Detecting Breast Cancer from Thermal Infrared Images by Asymmetry Analysis," Proceedings of the 22nd Annual International Conference of the IEEE: Engineering in Medicine and Biology Society, 2000. [1 1 ] L. F. Silva, D. C. M. Saade, G. 0. Sequeiros, A. C. Sivla, A. C. Paiva, R. S. Bravo, and A. Conci, "A New Database for Mastology Research with Infrared Image," Journal of Medical Imaging and Health Informatics, 2014.

[12] C.-R. Nicandro, M.-M. Efren, A. -A. Maria Yaneli, et al, "Evaluation of the Diagnostic Power of Thermography in Breast Cancer Using Bayesian Network

Classifiers," Computational and Mathematical Methods in Medicine, 2013.

[13] N. Arora, D. Martins, D. Ruggerio, et al, "Effectiveness of a noninvasive digital infrared thermal imaging system in the detection of breast cancer," The American Journal of Surgery, 2008.

[14] Y. R. Parisky, A. Sardi, R. Hamm, et al, "Efficacy of Computerized Infrared Imaging Analysis to Evaluate Mammographically Suspicious Lesions," American Journal of Roentgenology, 2013.

[15] H. R. B. Orlande, O. Fudyum, D. Maillet, and R. M. Cotta, Thermal

Measurements and Inverse Techniques, CRC Press, 201 1 .

[16] Haykin, S., Adaptive Filter Theory, Prentice Hall, 2002.

ADVANTAGES OF THE INVENTION

[0064] The present invention provides an improved method for detection of breast cancer. The method of the present invention improves sensitivity while at the same time preserving specificity in the detection of breast cancer. The method of the present invention is non-invasive and is low-cost and provides rapid and accurate results.

[0065] Methods according to the present invention possess industrial applicability for the diagnosis of breast cancer.

[0066] The method claims of the present invention provide specific method steps that are more than general applications of laws of nature and require that those practicing the method steps employ steps other than those conventionally known in the art, in addition to the specific applications of laws of nature recited or implied in the claims, and thus confine the scope of the claims to the specific applications recited therein.

[0067] The inventions illustratively described herein can suitably be practiced in the absence of any element or elements, limitation or limitations, not specifically disclosed herein. Thus, for example, the terms "comprising," "including," "containing," etc. shall be read expansively and without limitation. Additionally, the terms and expressions employed herein have been used as terms of description and not of limitation, and there is no intention in the use of such terms and expressions of excluding any equivalents of the future shown and described or any portion thereof, and it is recognized that various modifications are possible within the scope of the invention claimed. Thus, it should be understood that although the present invention has been specifically disclosed by preferred embodiments and optional features, modification and variation of the inventions herein disclosed can be resorted by those skilled in the art, and that such modifications and variations are considered to be within the scope of the inventions disclosed herein. The inventions have been described broadly and generically herein. Each of the narrower species and subgeneric groupings falling within the scope of the generic disclosure also form part of these inventions. This includes the generic description of each invention with a proviso or negative limitation removing any subject matter from the genus, regardless of whether or not the excised materials specifically resided therein.

[0068] In addition, where features or aspects of an invention are described in terms of the Markush group, those schooled in the art will recognize that the invention is also thereby described in terms of any individual member or subgroup of members of the Markush group. It is also to be understood that the above description is intended to be illustrative and not restrictive. Many embodiments will be apparent to those of in the art upon reviewing the above description. The scope of the invention should therefore, be determined not with reference to the above description, but should instead be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled. The disclosures of all articles and references, including patents and patent publications, are incorporated herein by reference.