Breast cancer affects one in every eight women in the United States and is ranked second only to lung cancer in cancer-related deaths among women. Randomized controlled trials and large-scale screening programs have shown that periodic screening leads to earlier detection and a reduction in breast cancer mortality [1,2,3,4]. Currently, self-examination and regular mammography are the most effective techniques for detecting breast cancer, and it is recommended that women begin undergoing regular screening mammography during their 40s [5, 6]. Mammography, the gold standard screening modality, reveals occult malignant lesions in asymptomatic women at an earlier stage and in smaller lesions, generally producing a more favorable prognosis than is possible by self-examination.

Despite the value of mammography in revealing breast malignancies, most radiographically identified lesions are ultimately found to be benign on histologic assessment after biopsy. National statistics indicate that between two thirds and four fifths of all breast biopsies have a benign outcome [7,8,9]. Breast biopsy costs vary considerably, with the expenses ranging from less than a thousand dollars to several thousand dollars [10,11,12]. Thus, noninvasive methods are needed to aid clinicians in distinguishing benign from malignant breast tissue.

Imaging techniques used in conjunction with mammography and physical examination often include sonography, CT, and MR imaging. Although promising, CT and MR imaging are generally considered too costly for routine use. In addition, MR imaging has not proven to be efficacious in young patients, because MR findings may lead to an unacceptably high number of workups for benign lesions while detecting few malignancies [13]. Sestamibi scintimammography has also been used with varying success to evaluate breast lesions scheduled for biopsy: sensitivities have ranged from 55% to 98%, and specificities have ranged from 79% to 95% [14,15,16]. Disadvantages of this technique include the expense, duration of the examination, and patient exposure to ionizing radiation. Sonography is more cost-effective than either CT or MR imaging but has limitations [17,18,19,20]. Like mammography, sonography relies on the expertise of the technician and the interpretative skills of the radiologist.

One promising modality for aiding clinicians in differentiating malignant from benign breast lesions is infrared imaging. This modality is noninvasive and detects physiologic tissue response, rather than evaluating anatomic features. As heat is released from the body, a portion is in the form of infrared radiation. Several physiologic features related to malignant tissue may contribute to the infrared signal, including increased blood flow in the area surrounding a malignancy, angiogenesis, and the release of vasoactive mediators [21,22,23,24,25,26]. The infrared imaging system uses a camera that is highly sensitive to infrared radiation in the appropriate spectrum. This computerized system is designed to show that benign tissue can be differentiated from neoplastic tissue on the basis of the relatively higher strength of the infrared signal in malignant tissue.

In this study, we performed infrared imaging using a computerized dynamic imaging system to capture a series of sequential images of the breast over a short time, which included a period when cool air was blown over the breast. The sequential images were processed using complex proprietary algorithms that assess the infrared information obtained in an identified suspicious area.

The data presented here were collected during a 4-year clinical trial using infrared imaging of patients for whom breast biopsy was recommended on the basis of either clinical or mammographic findings. During the entire course of the study, more than 2400 subjects underwent infrared imaging. The subjects were recruited from five sites that are located in Los Angeles; in Baltimore; in Washington, DC; near Boston; and in Miami. During the initial phases of the study, the subjects' pathology results from biopsies were available to develop the algorithms on which the proprietary analytic software was based.

Subsequent subject data about the biopsy results were collected so that neither the investigational study sponsor nor the evaluating physicians were aware of the outcome, and these data were used to test the efficacy of the infrared imaging system. The results discussed here are from 769 subjects enrolled in this phase of the clinical investigation. During this phase, the evaluating mammographers used mammograms to localize the suspicious lesion on an infrared image but did not have access to the biopsy outcome. The primary goal was to determine the efficacy of dynamic computerized infrared imaging in enabling clinicians to distinguish benign from malignant lesions. In addition, various breast and lesion features, including breast density, tumor size, and lesion appearance, were evaluated to determine their potential effect on or their correlation to dynamic infrared imaging performance.

Of the 769 subjects evaluated, 766 were women and three were men. The racial composition of the study group reflected that of the United States population: white (n = 463), black (n = 207), Hispanic American (n = 81), Asian American (n = 13), and others (n = 5). The age range of the study subjects extended through all the age categories at risk for breast cancer. Most of the subjects (n = 433) were between the ages of 40 and 60 years at the time of infrared imaging. Of the remaining subjects, 68 were younger than 40 years and 268 subjects were older than 60 years. Of the subjects included in the primary efficacy group, 90 had more than one lesion biopsied; most of these subjects (n = 77) had two lesions excised. The remaining 13 subjects had more than two lesions biopsied. The total number of evaluated lesions equaled 875. Of these lesions, 187 were malignant and 688 were benign, so 78.6% of the biopsied lesions were benign. This finding is in accord with national statistics [7,8,9].

The data analysis used the index-of-suspicion threshold to determine whether an index-of-suspicion result was negative or positive for malignancy. The infrared results were established before the pathology outcomes were disclosed. As shown in Table 1, the 875 evaluated lesions resulted in 2299 valid index-of-suspicion determinations by the assigned radiologists. The infrared imaging—based index of suspicion yielded a 97% sensitivity (95% CI, 94-98%) and a negative predictive value of 95% (95% CI, 91-98%). The specificity was 14% (95% CI, 12-16%), and the positive predictive value was 24% (95% CI, 23-24%). It is important to note that all these subjects under-went biopsy because the original site physician characterized a lesion as suspicious, so the standard clinical workup resulted in a sensitivity of 100% and a specificity of 0%.

The negative predictive value of 95% resulted from evaluators obtaining 13 negative infrared results for seven malignant lesions in six subjects, as shown in Table 2. The interob-server variability among the seven evaluators was as follows: the first evaluator had four of the 13 false-negative results, the second had three false-negatives, the third had two false-negatives, and the remaining four evaluators had one false-negative result each.

Factors that may have contributed to the 13 false-negative findings are as follows. Evaluators may have been confused about the location of the mammographically identified and biopsied lesion, which possibly led to localization of the wrong lesions on the infrared images, because three of the six subjects (cases 1, 2, and 3) had multiple biopsied lesions (lesions were assigned an alphabetic designation as illustrated in Table 2). One of the 41-year-old subjects (case 1) had two additional malignant lesions biopsied, and both of these lesions were correctly evaluated as positive for malignancy by the infrared results. The other 41-year-old subject (case 3) also had three areas biopsied with benign outcomes in lesions A and C. In addition, the evaluators noted that three of the six subjects (cases 1, 3, and 6) appeared to have been poorly positioned during infrared imaging, which also may have adversely affected the outcomes. Finally and perhaps of greatest significance, all the missed lesions were microcalcifications. Pathology findings revealed that four of the seven lesions were ductal carcinoma in situ, whereas two cases were described as ductal carcinoma in situ with stage I focal microinvasion. The one exception was lesion A in case 1, described by the physician at the original site as an asymmetric density containing microcalcifications. The pathology result for this lesion indicated intraductal and infiltrating poorly differentiated ductal carcinoma. This case was problematic for a variety of reasons: the subject had multiple lesions, the positioning of the breast for infrared analysis was poor, and a review of this case revealed that the region-of-interest placement was questionable.

Because all the lesions to which evaluators assigned false-negative results were microcalcifications, we analyzed the subset of lesions that excluded those described as microcalcifications alone. This subset consisted of 479 lesions, of which 110 were malignant and 369 benign, in 448 subjects. Table 3 shows the infrared results for this restricted subset. The infrared findings had resulted in a sensitivity and negative predictive value of more than 99% (95% CI, 96-100%) with 291 true-positive results and one false-negative result. The false-negative result in this subset was for case 1 (Table 2), which we described earlier in this article. Although all three evaluators attempted to assess this lesion on infrared imaging, two of the three did not think that the quality of the infrared image was acceptable for analysis. When the same subset was restricted further to include only the cases for which all three evaluators had completed an assessment, the sensitivity and negative predictive value were 100%, the specificity was 19%, and the positive predictive value was 27%.

Table 4 illustrates the distribution of pathology results that were found in the 110 malignant and 369 benign cases. The largest number of malignant results, comprising 68 cases, was in the invasive category: ductal carcinoma alone or ductal carcinoma in combination with other pathology. In the benign category, the most common findings were fibroadenoma and fibrocystic disease, which included 226 cases. These findings accounted for approximately 60% of the benign outcomes. Another 24 cases with benign outcomes were in the high-risk category (i.e., atypical hyperplasia or phyllodes tumor).

Among the features noted during the clinical trial and evaluation phase were breast composition and lesion size. Breast composition appeared to influence the infrared imaging performance. For all 875 lesions, specificity was better for subjects with extremely dense breast tissue than for those with fatty breast tissue (Fig. 2). Infrared results for lesions in breasts with a scattered fibroglandular density and in heterogeneously dense breasts resulted in intermediate specificity values. Sensitivity, which ranged from 94% to 100%, did not appear to be significantly affected by breast density; the lesions with false-negative infrared results described in Table 2 occurred in breasts of scattered fibroglandular density or that were heterogeneously dense). The relationship of breast density and specificity of infrared results became more pronounced when evaluating lesions that were described only as masses. In this subset, which included 412 masses with 90 malignancies, the sensitivity was 100% and the specificity increased from 8% in almost entirely fatty breast tissue, to 15% in scattered fibroglandular tissue, to 20% in heterogeneously dense tissue, and finally to 26% in extremely dense breast tissue. In addition, lesion size of malignant masses correlated well with the index-of-suspicion infrared value (Fig. 3).

More than 1 million breast biopsies are performed every year in the United States and approximately 75-80% yield a benign outcome [7,8,9]. The percentage of breast biopsies with a benign result in this clinical trial (78.6% [688/875 lesions]) was consistent with the national statistic; similar percentages were found among the subsets of lesions analyzed. Mammography is the well-established gold standard for identifying suspicious lesions on the basis of anatomic information. However, as radiologists assess the various ways to improve detection of breast cancer and to better distinguish between benign and malignant suspicious lesions, increasing interest is being focused on the physiologic profile of the disease. Other diagnostic modalities for breast cancer that rely, at least in part, on physiologic processes include sestamibi scintimammography, Doppler sonography, gadolinium-enhanced MR imaging, and positron emission tomography.

The computerized infrared imaging system that we used was developed to assist physicians in differentiating benign tissue from malignant tissue by characterizing different patterns in the infrared signal emitted by the breast. The infrared imaging system could be used as an adjunct for further evaluating a mammographically apparent breast abnormality when the radiologist has a low-to-moderate suspicion that a malignancy is present. This modality is not designed to be a screening tool for identifying or localizing malignancies or to delay biopsy of highly suspicious lesions. Therefore, a high negative predictive value is essential, whereas the positive predictive value does not have as great a clinical utility.

Although this clinical trial was designed to evaluate the adjunctive value of infrared imaging to mammography, the radiology reports revealed that sonography was also performed in a significant number of patients (approximately 45%). This finding suggests that the infrared imaging assessment would have adjunctive value to standard clinical practice when both mammography and sonography are commonly used in the determination of whether to recommend biopsy.

Our results suggest that infrared imaging is most efficacious with lesions that are masses. A high negative predictive value was found when assessing these lesions (99%). In contrast, when all the lesions—including those described as microcalcifications—were assessed, the overall negative predictive value was 95%. The malignancies assigned a negative infrared test result by at least one evaluator included seven lesions that contained microcalcifications. Pathologic examination revealed that six of the seven were low-grade ductal carcinoma in situ (n = 4) or ductal carcinoma with microinvasion (n = 2); the seventh was ductal carcinoma in situ with infiltrating poorly differentiated ductal carcinoma.

Several physiologic factors might explain why infrared imaging did not perform optimally for assessment of ductal carcinoma in situ. A malignant lesion confined to the basement membrane may not substantially influence the physical milieu, whereas a more invasive lesion would. Another factor may be that angiogenesis is less advanced during the earlier stages of ductal carcinoma in situ when the tumor is still sequestered in the duct. Additional studies are needed to address these issues. In our study, we found that most of the lesions containing microcalcifications were correctly determined to be malignant by the infrared imaging procedure.

Of prime importance in assessing the clinical utility of a new modality involved in the diagnosis of breast cancer is the role that the technology might assume in clinical practice. Undergoing breast biopsy of benign tissues places substantial physical and psychologic burdens on the patient without improving patient care. In addition, such testing has a great economic impact on the health care system [10,11,12]. Infrared imaging assessment offers a noninvasive, safe procedure that could be helpful in determining whether an immediate biopsy is warranted.

As clinicians acquire experience with infrared technology, they may wish to continue to follow up patients to validate a benign result. One possible scenario is that a negative infrared result for a suspicious lesion could be considered similar to a mammographically detected lesion for which short-interval follow-up is recommended—that is, a BI-RADS (Breast Imaging Reporting and Data System [27]) category 3 lesion. Generally, short-interval follow-up means reassessing the patient by both clinical and radiologic examinations in 3-6 months instead of performing the standard screening annually [27,28,29,30]. Factors supporting the value of the short-interval follow-up include lower levels of patient stress, increased subsequent compliance with mammography screening, and lower overall cost [31,32]. The infrared imaging technology described here is currently being reviewed by the United States Food and Drug Administration, and that agency, as well as clinical experience, will ultimately dictate the appropriate use of this technology.

In summary, infrared imaging holds great promise in the management of breast lesions that can be localized on mammography, either by direct visualization or with commonly used markers. Infrared imaging is an economic and safe modality that provides physiologic data about a lesion. The physiologic view provided by infrared imaging complements the anatomic view provided by mammography with a very high sensitivity and negative predictive value in masses. Thus, this dynamic computerized infrared imaging system could be a valuable addition to the physicians' armamentarium of diagnostic tools.

Supported by Computerized Thermal Imaging, Lake Oswego, OR.

Address correspondence to Y. R. Parisky.