A Prospective Study on Diagnostic Accuracy of Breast Thermography over Conventional Breast Imaging (Mammography and Supplemental Ultrasonography) in Average-Risk Indian Women

Abstract

Background

The current recommendation for breast cancer screening is mammography. However, it has a high false negative rate, is less sensitive in dense breasts, and is associated with radiation exposure to the breast. Thermography measures body surface temperatures. In conventional thermography, the patient sits in front of the camera. Infrared images of the breast are captured in three views. Rotational thermography (Illumina360°) images with 360° views of one breast are obtained at a time in two different controlled temperatures when it is freely suspended.

Objective

Primary: Descriptive study of thermographic images and correlation with routine conventional imaging (mammography and supplementary ultrasonography). Secondary: To assess the diagnostic accuracy of thermography with conventional work up as the gold standard.

Materials and Methods

It is a prospective observational study. Inclusion criteria are positive finding on mammography and age 18 to 75 years. Patients who cannot lie in the prone position, patients who are unable to follow instructions, patients with fever, and pregnant and lactating women were excluded. Setting: tertiary cancer care center. Patients with unilateral positive mammography and contralateral negative mammography, where available, underwent thermography and findings from both modalities were compared. Sample size: 100 patients (198 breasts).

Results

The sensitivity of thermography in comparison to mammography was 83.87%, with specificity 10.81%, diagnostic accuracy 56.57%, positive predictive value 61.18%, and negative predictive value 28.57%.

Conclusion

Thermography has high sensitivity and low specificity with a high false positive rate and thus a low diagnostic accuracy. It could not reliably differentiate benign from malignant lesions based on temperature risk stratification. Mammographic breast density and menstrual status did not have an effect on its ability to pick up lesions. However, specificity and diagnostic accuracy in premenopausal women imaged in the first half of menstrual cycle were more than those in the second half. If breast thermography is explored as a screening modality for early detection of breast cancer, its limitations can be a lower detection rate of smaller cancers and false positive uptake in high proportions of normal breasts.

Keywords

breast thermography - ultrasonography - mammography

Introduction

Breast cancer is the most common cancer among women in India, with the highest mortality of 11.4% among all cancers and 24% of cancers in women.[1] Detecting breast cancer early and downstaging disease at presentation have a profound impact on disease-free survival,[2] rendering a need toward developing reliable and economically viable techniques for early as well as accurate detection. Many techniques are available for breast imaging, of which, the current recommendation for screening and diagnosis of asymptomatic breast cancer is mammography (MG); however, it has a high false negative rate, is less sensitive in dense breasts, and is associated with radiation exposure.[2] [3] [4] [5] Globally, only a few countries implement a National Screening Mammography program, primarily due to societal and economic challenges. Ultrasonography (US), a lesser expensive alternative, has been explored as a screening tool with high sensitivity but poorer specificity, and it is heavily observer-dependent.[6] Magnetic resonance imaging (MRI), too, is an expensive and relatively inaccessible technique with a high false positive rate.[7] A need is evident for a less expensive, easily accessible, and robust imaging technique with superior diagnostic accuracy (DA). It is this lacuna that breast thermography (BT), based on infrared temperature-based imaging, is targeting to fill and is explored as a potential tool in breast cancer screening.

BT provides a pictorial representation of surface temperatures of the breast, depending on its emitted infrared radiation.[8] It is based on the principle that chemical and blood vessel activity in areas associated with breast cancer is higher, leading to increased surface temperature.[9] A review of the literature shows variable results of BT, both as a stand-alone tool and as an adjunct to existing imaging techniques, and we thereby proposed to conduct a pilot study in the Indian population.

Materials and Methods

An institutional ethics committee–approved prospective observational study was conducted in a tertiary referral cancer center. Hospital-registered women after initial clinical assessment are referred to the Department of Radiology for a mammogram examination. Consecutive patients in the age group of 18 and 75 years with positive findings on MG who consented to the study were included in the study. Women who were unable to lie down in the prone position, were febrile, pregnant or lactating, and those who did not consent were excluded. The study continued until 100 women consented for the study, of whom 2 women had unilateral mastectomy. The sample size thus constituted 198 breasts. The study design was as follows ([Fig. 1]).

Fig. 1 Algorithm for Accrual.

BT was performed on the CURA rotational BTILLUMINA360 machine; mammograms were acquired using the Senographe DS (General Electric) and Selenia dimensions (Hologic); and US was performed on the Voluson 730 proBT05 (General Electric). Data acquisition of both BT and MG of patients was performed on the same day. The procedure required the women to lie comfortably in a prone position on the ILLUMINA360° unit with one breast freely suspended into a chamber through a small circular aperture. The prone position ensures comfort and reduces artifacts that may arise due to patient movement. A robotic arm rotates clockwise around the suspended breast and captures images at spatial intervals of 15 degrees, resulting in a series of 24 images. Both breasts are imaged sequentially. As a protocol, the left breast is evaluated first, followed by the right, at the chamber's ambient temperature, forming a pre-cool series of 24 images. Then the temperature inside the chamber is reduced by 3 to 4°C and imaging is resumed, generating a set of 24 images referred to as post-cool series. The robotic arm also moves in a direction to image the nipple region, forming the frontal image. Multiple-view captures ensure temperature recording of the entire breast surface. It takes 14 to 15 minutes for the entire process of BT. BT readings are interpreted as temperature values in °C and classified as a risk score into high, moderate, and low risk.[10] Thermal frames of interest are marked based on the above criteria. A focal lesion on BT is identified as a focal area of increased temperature that is discontinuous from the chest wall. The post-cool images are read because of the enhanced temperature differences between the normal and abnormal regions in the breast.[10] The mammograms were read by a breast radiologist and the thermography was read by a radiologist and a thermography technologist assistant. Standard-of-care evaluation of the breast on MG and US interpretation is based on the ACR BI-RADS (American College of Radiology Breast Imaging Reporting and Data Systems) fifth edition, and findings were classified as BI-RADS 0 (incomplete assessment), 1 (negative), 2 (benign), 3 (probably benign), 4 (suspicious for malignancy), 5 (highly suggestive of malignancy), and 6 (known biopsy-proven malignancy).[11] In our study, the above-mentioned standard-of-care conventional imaging (MG and US) was taken as the gold standard and formed the basis of triaging lesions for biopsy correlation. Variations in types of uptake and conventional image correlation may yield true positive ([Fig. 2]), false positive ([Fig. 3]), true negative ([Fig. 4]), and false negative cases ([Fig. 5]) for both benign and malignant conditions.

Fig. 2 (A), (B) Right mammogram of a 49 year old post-menopausal woman shows a round high density mass with circumscribed margins in the upper inner quadrant of the right breast Fig part (C). On USG, it corresponds to a hypoechoic mass with microlobulated margins at 2-3 o clock position in the right breast. (D) On thermography it corresponds to a lesion with raised temperature (33.67°C) in 2-3 o clock position with surrounding area of graded decrease in temperature in the right breast, rendering a True Positive result.

Fig. 3 (A), (B) Right mammogram of a 57 year old post- menopausal woman with predominantly fatty parenchyma shows no abnormality. (C)On thermography an area of raised temperature is seen in the lower quadrants of right breast rendering a False Positive result.

Fig. 4 (A), (B) Left mammogram of a 48 year old premenopausal woman shows heterogeneously dense fibroglandular parenchyma with no significant abnormality. (C) On thermography, left breast shows graded increase in temperature from the nipple to the chest wall with no focal area of raised temperature, rendering a True Negative result.

Fig. 5 (A), (B) Right mammogram of a 57 year old woman with non-dense breast shows a high density irregular mass with spiculated margins in the upper outer quadrant. (C) On screening USG, it corresponds to a cluster of hypoechoic lesions with angular margins measuring approximately 3.6 x 2.3 cm in size at 11-9 o clock position. (D) Thermography shows uniform low temperature with no abnormal uptake. On histopathology the lesion was a grade 3 invasive ductal carcinoma. This classifies as False Negative result.

Results

Demographics with age distribution, menstrual status, and distribution of types of breast parenchymal density are given in [Table 1]. Histopathological types of breast malignancies detected in this cohort and their distribution are summarized in [Table 2].

Table 1 Demographics

	Frequency	Percentage
Age range
21–35 years	11	11%
36–50 years	49	49%
51–75 years	40	40%
Menstrual status
Premenopausal	29	29
Peri-menopausal  (less than 1 year of cessation of menstruation)	4	4
Post-menopausal	53	53
Post-hysterectomy	14	14
Breast parenchymal density
Predominantly fatty parenchyma	4	4
Scattered fibroglandular parenchyma	22	22
Heterogeneously dense fibroglandular parenchyma	62	62
Extremely dense fibroglandular parenchyma	12	12
BI-RADS
1	74	37.4%
2	22	11.1%
3	4	2%
4a	6	3%
4b	7	3.6%
4c	29	14.6%
5	56	28.3%

Abbreviation: BI-RADS, Breast Imaging Reporting and Data System.

Table 2 Type of malignancies detected

Type of malignancy	Number
Invasive breast ca, no special type, Grade 3	68
Invasive breast ca, no special type, Grade 2	8
Invasive metaplastic ca, Grade 3	1
Invasive pleomorphic lobular carcinoma, Grade 3	1
Micropapillary ca (FNAC)	1
Ductal carcinoma in situ	1
Invasive lobular carcinoma, Grade 2	1
Scanty tubular carcinoma, Grade 2	1
Spindle cell tumor (sarcoma) Mets to breast	1
Total	83

Abbreviation: FNAC, fine needle aspiration cytology.

Diagnostic Accuracy of BT Compared with MG for Lesion Detection

Of the total 198 breasts examined, 124 cases were positive for findings on MG and US imaging. Of these 124 cases, 104 correctly correlated with BT as true positive cases ([Figs. 6] [7] [8]), while 20 cases had no correlation, rendering BT as a false negative. Of the 74 cases that were negative on MG and US, only 8 were concordant with BT as true negatives, while 66 were falsely identified to have abnormal temperature on BT, representing false positive cases. This attributes to a reasonable sensitivity of BT at 83.87% (95% confidence interval [CI]: 76.19–89.86%), a very low specificity of 10.81% (95% CI: 4.78–20.20%), a positive likelihood ratio of 0.94 (95% CI: 0.84–1.05), a negative likelihood ratio of 1.49 (95% CI: 0.69–3.21), a positive predictive value of 61.18%, and a negative predictive value of 28.57%, with a poor DA of 56.57%.

Fig. 6 (A), (B) A 36 year old premenopausal woman presents with a palpable left breast lump since 3 years. Left mammogram shows a dense breast with a high density mass with obscured margins in the upper outer quadrant. (C) On USG, it corresponds to an oval, regular, isoechoic mass with internal cystic changes, with parallel orientation at 1-2 o clock position. Lesion classified as BI-RADS assessment category of ([Figure 4A]). (D) On thermography, it corresponds to a high temperature lesion at 2 o clock position in the left breast. On histopathology, it rpresented fibrocystic change. This represents concordance of BT with imaging for detection, classifying as True positive.

Fig. 7 (A), (B) Left mammogram of a 33 year old premenopausal woman shows dense breast with a high density irregular mass with obscured margins in the inner central region of the left breast. (C) On USG it corresponds to a hypoechoic lesion measuring 1.4 x 1 cm in the retroareolar position, with echogenic halo and focal angular margin. This classifies as BIRADS ([Figure 4B]). (D) On thermography, it corresponds to a high temperature lesion (30.3 degree C) in the retroareolar location. On Histopathology, the lesion was grade III invasive ductal carcinoma, and the case is rendered as True Positive result.

Fig. 8 (A), (B) Left mammogram of a 57 year old post menopausal woman shows a high-density mass in the upper outer quadrant with pleomorphic micro-calcifications. n additional small isodense lesion is seen in the upper inner quadrant. (C), (D), (E) On USG correlation, a hypoechoic mass with spiculated margins is seen at 2 o clock position. Additional satellite lesion is seen at 11 o clock position in the left breast. Findings classified as BIRADS 5. (F) On thermography, the index mass is seen as an area of raised temperature at 2-3 o clock position. (G) The additional lesion is also seen as an area of raised temperature in the left breast at 11 o clock position. On Histopathology, the lesion was a grade II invasive breast carcinoma. This renders as a True Positive result.

Performance of BT on Various Features

• Correlation of BT uptake compared with MG based on BI-RADS.

  • – Of 124 positive cases on MG and US ([Fig. 9]), 22 were classified as BI-RADS 2. Of these, BT showed a thermal uptake in 13 cases (59%). One of four cases under BI-RADS 3 showed thermal uptake. Of the 98 breasts categorized as BI-RADS 4 or 5, thermal uptake was seen in 90 (91.84%) cases at the site of the disease identified on MG, while in 8 (8.16%) cases, the thermal uptake was seen as a possible spurious uptake in another quadrant, distinct from the obvious site of disease and considered as discordant.

• Correlation of BT risk stratification with BI-RADS assessment.

  • – [Table 3] shows the correlation of thermographic risk stratification and the BI-RADS assessment of cases on MG. Breasts with high thermographic risk stratification had a higher number of normal and BI-RADS 5 lesions (31%) as compared with moderate (20%) and low-risk (0%) stratification lesions. All breasts classified as normal on thermographic risk stratification belonged to the BI-RADS 1 category. However, in high thermographic risk stratification breasts, 33% were BI-RADS 1; 31% were BI-RADS 5; 15% were BI-RADS 4c; 11% were BI-RADS 2, 4% were BI-RADS 4b, 4% were BI-RADS 4a, and 2% were BI-RADS 3. There was no significant correlation between thermographic risk stratification and BI-RADS, with a p-value of 0.275.

• Correlation of BT with histopathology.

  • – [Table 4] documents that malignant lesions showed variably elevated temperatures, with 94% exhibiting high-risk stratification, 4.8% moderate risk stratification, and 1.2% low risk stratification. However, benign lesions also showed abnormal temperatures, with 84.6% exhibiting high-risk stratification and 15.4% low-risk stratification ([Fig. 10]).

• Correlation of BT concordance with breast density.

  • – [Table 5] shows increasing concordance of BT with increasing breast density, but no significant correlation (p-value is 0.299) of the breast density seen with the correct identification of the breast being normal or abnormal.

• Correlation of BT concordance with lesion size.

  • – [Table 6] shows that no significant correlation was found with sizes among benign lesions, whereas increasing concordance was observed with increasing sizes in suspicious lesions, such that larger lesions were detected on BT with a significant p-value of 0.0256.

• Correlation of BT concordance with menopausal status.

  • – [Table 7] shows that no effect of the thermographic correlation was observed with menopausal status (p-value 0.49).

• Correlation of BT concordance with phase of menstrual cycle in premenopausal women.

  • – Out of 100 patients, 22 of 29 premenopausal women had regular menstrual cycles and remembered their last menstrual period date. Twelve of these were in the first half of the menstrual cycle and 10 in the second half. [Table 8] shows that there was greater concordance with BT in the first half, but with a nonsignificant p-value of 0.262.

  • – Twenty-four breasts were evaluated in the first half of menstrual cycle. Of the normal 7 breasts, BT truly identified 2, with false positive uptake in 5 breasts. Of the 17 breasts with findings, BT truly identified 14 breasts with thermal uptake, while 3 did not show a thermal uptake, resulting in a sensitivity of 82.35%, a specificity of 28.57%, and a DA of 66.67%.

  • – Twenty breasts were evaluated in the second half of menstrual cycle. Of the normal 9 breasts, BT truly identified 1, with false positive uptake in 8 breasts. Of the 11 breasts with findings, BT truly identified 9 breasts with thermal uptake, while 2 did not show a thermal uptake, resulting in a sensitivity of 81.82%, a specificity of 11.11%, and a DA of 50% ([Table 9]).

  • – Overall, a greater increase in the specificity from 11.11 to 28.57% with an increase in the DA from 50 to 66.67% was observed in the first half of menstrual cycle.

Fig. 9 Breast thermography correlation with Final BI-RADS Assessment Category.

Table 3 Comparison between BI-RADS categories and BT risk stratification

	BI-RADS on MG
Thermographic risk stratification		1	2	3	4a	4b	4c	5
	Normal	4	0	0	0	0	0	0
	Low	5	2	1	0	1	1	0
	Moderate	9	1	0	0	0	2	3
	High	56	19	3	6	6	26	53

Abbreviations: BI-RADS, Breast Imaging Reporting and Data System; BT, breast thermography; MG, mammography.

Table 4 Histopathology and thermographic correlation

Histopathology and thermographic temperature correlation
	High	Moderate	Low
Benign	11 (84.6%)	0	2 (15.4%)
Phyllodes	1 (50%)	0	1 (50%)
Malignant	78 (94%)	4 (4.8%)	1 (1.2%)

Fig. 10 Breast thermography correlation with Histopathology.

Table 5 BT correlation and breast density

	Thermographic correlation
Breast density	Total	Concordance	Discordance
Extremely dense fibroglandular parenchyma	24	15 (62.5%)	9 (37.5%)
Heterogeneously dense fibroglandular parenchyma	123	71 (57.72%)	52 (42.28%)
Scattered dense fibroglandular parenchyma	43	24 (55.81%)	19 (44.19%)
Predominantly fatty parenchyma	8	2 (25%)	6 (75%)

Table 6 Thermographic correlation based on the size of the lesion

MG categories	BT correlation	Size of the lesion
		Total	<2 cm	2–5 cm	>5 cm
Benign	Concordance	12	10 (66.7%)	2 (50%)	0
	Discordance	7	5 (33.3)	2 (50%)	0
Probably benign	Concordance	1	1 (25%)	0	0
	Discordance	3	3 (75%)	0	0
Malignant	Concordance	90	12 (75%)	47 (96%)	31 (93.9%)
	Discordance	8	4 (25%)	2 (4%)	2 (6.1%)

Abbreviations: BT, breast thermography; MG, mammography.

Table 7 Thermographic correlation with menopausal status

Menstrual status	Thermographic correlation
	Total	Concordance	Discordance
Premenopausal	58	36 (62.07%)	22 (37.93%)
Peri-menopausal	8	3 (37.5%)	5 (62.5%)
Post-menopausal	104	56 (53.85%)	48 (46.15%)
Post-hysterectomy	28	17 (60.71%)	11 (39.29%)

Table 8 Phase of the menstrual cycle and thermographic correlation

	Phase of menstrual cycle
BT correlation as per menstrual cycle	Day 3–14	Day 15–2
Concordance	16 breasts (66.7%)	10 breasts (50%)
Discordance	8 breasts (33.3%)	10 breasts (50%)

Abbreviation: BT, breast thermography.

Table 9 Diagnostic accuracy for BT for premenopausal women imaged in the first half (Day 3–14) and second half (Day 15–2) of the menstrual cycle

	Statistical values		95% CI
Phase of menstrual cycle	First half	Second half	First half	Second half
Sensitivity	82.35%	81.82%	56.57–96.20%	48.22–97.72%
Specificity	28.57%	11.11%	3.67–70.96%	0.28–48.25%
Positive likelihood ratio	1.15	0.92	0.69–1.93	0.64–1.32
Negative likelihood ratio	0.62	1.64	0.13–2.93	0.18–15.26
Positive predictive value	73.68%	52.94%	62.53–82.45%	43.93–61.77%
Negative predictive value	40%	33.33%	12.31–75.99%	5.09–82.34%
Accuracy	66.67%	50%	44.68–84.37%	27.2–72.8%

Abbreviations: BT, breast thermography; CI, confidence interval.

BT Changes in Normal Breasts Based on Phase of Menstrual Cycle

Of the 74 normal breasts, 16 premenopausal patients (16 breasts) had regular menstrual cycles and remembered their last menstrual period, and were observed to have greater thermal uptake and hence greater false positives in the second half of menstrual cycle with a nonstatistically significant p-value of 0.333 ([Table 10]).

Table 10 Temperature risk variations according to the phase of the menstrual cycle in normal breasts

	Day 3–14 (total: 7 breasts)	Day 15–2 (total: 9 breasts)
Normal	2 (28.6%)	0 (0%)
Low	1 (14.3%)	1 (11.1%)
Moderate	1 (14.3%)	1 (11.1%)
High	3 (42.8%)	7 (77.8%)

BT and Clinical Examination

[Fig. 11] shows the correlation of clinical examination findings and thermography. It depicts the relationship between clinical palpability, MG findings, and thermography results, showing how many breasts were positive or negative on thermography.

Fig. 11 Breast thermography correlation with Clinical Examination.

[Fig. 12] shows the correlation of clinical examination findings with thermography risk stratification. The figure shows how palpable and nonpalpable breast abnormalities on MG are further categorized into high, moderate, or low risk based on thermography.

Fig. 12 Clinical examination and Thermography risk stratification correlation, in Abnormal Mammograms.

Follow-Up Assessment Results

Of the 100 patients included in the study, 31 patients followed up at different time intervals up to 6 months post-BT. Twenty-eight of these patients were following up on neoadjuvant chemotherapy, of which 20 underwent ipsilateral MG and 8 underwent bilateral breast mammograms. One patient underwent modified radical mastectomy for breast cancer, and hence a contralateral normal follow-up mammogram was performed. One patient with post-excision biopsy underwent a bilateral mammogram. One patient had defaulted and underwent a repeat bilateral breast mammogram. Among these patients on follow-up, 9 had false-positive thermal uptake on BT, but none showed a new lesion on conventional MG imaging.

Discussion

The sensitivity of BT in comparison to MG in our study was 83.87% (95% CI: 76.19–89.86%) and the specificity was 10.81% (95% CI: 4.78–20.20%). The overall DA was 56.57% (95% CI: 49.35–63.58%). The positive predictive value was 61.18% (95% CI: 58.52–63.77%) and the negative predictive value was 28.57% (95% CI: 15.66–46.29%).

A direct comparison of studies on BT is difficult as they have used different technologies of BT with differing regulatory compliances and varying results. A study conducted by Parisky et al[12] in the year 2003 demonstrated a sensitivity of 97% (95% CI: 94–98%) and a specificity of 14% (95% CI: 12–16%), similar to our study, with a positive predictive value of 24% (95% CI: 23–24%) and a negative predictive value of 95% (95% CI: 91–98%). They conducted the study with a similar principle of imaging to ours, using the dynamic computerized infrared imaging system BCS2100 (Computerized Thermal Imaging, Ogden, Utah). It included patients planned for biopsy, excluded patients in whom findings of BT and MG did not match, and excluded normal breasts, resulting in selection bias. A study conducted by Omranipour et al[13] in the year 2016 included 132 patients planned for biopsy, also a biased selection. The sensitivity and specificity of BT in this study were 81 and 57% as compared with 80 and 73%, respectively, for MG. The sensitivity of BT and MG combined was 96%; however, the specificity and DA dropped to 44 and 70%, respectively. Another study conducted by Arora et al[14] in the year 2008 had a sensitivity of BT up to 96.7% and a specificity up to 44.1%. Williams[15] showed a higher specificity of 74% compared with MG, but a lower sensitivity of 61% in his study.

Detection of false uptake in normal breasts is an important consideration when evaluating the utility of BT as a screening tool. Feig et al's study,[8] conducted in the year 1977, found a large number of positive thermographic findings in the general population. Our study similarly found more number of positive thermograms in BI-RADS 1 cases proven on mammogram and supplementary US. Out of the 74 normal breasts, BT correctly identified only 8 breasts to be normal (10.8%) and in 66 breasts (89.19%) BT falsely identified a lesion, thus being falsely positive.

BT shows better detection for larger lesions. Of the 98 malignant lesions analyzed in our study, 31 cases (93.9%) were more than 5 cm and had a positive correlation with BT, and a negative correlation in 2 cases (6.1%). Forty-seven cases (96%) of size 2 to 5 cm had positive and two cases (4%) had negative correlation. Twelve cases (75%) of size less than 2 cm had positive and four cases (25%) had negative correlation. This relation of size and thermographic correlation for malignant lesions was significant with a p-value of 0.0255 (p < 0.05). The size correlation has a reverse significance, as larger tumors are often palpable, but the challenge remains in impalpable and smaller sizes for early detection, where BT does not seem reliable. Similar findings were observed in a study conducted by Feig et al, which included 16,000 patients and showed that larger lesions were better identified on BT.[8] In 2016, Faustino-Rocha et al[16] conducted a study in 2,036 patients, of whom 480 patients were proven to have malignancies. The sensitivity and specificity of infrared BT were superior to those of MG and US in lesions less than 2 cm in diameter, and MG had a better DA only in lesions larger than 2 cm in diameter. However, this study primarily focused on thermography's ability to detect lesions based on size, without clearly differentiating its performance in distinguishing malignant from benign lesions. In a study conducted in rat mammary tumors, it was found that the size of the tumor volume correlated with the maximum tumor temperature.[16] Sterns et al conducted a study in 420 patients with invasive breast cancer using a liquid crystal contact BT and found that an abnormal thermogram is associated with a large tumor size, a high grade, and a lymph node positivity. However, the study also concluded that BT was not an independent prognostic indicator.[17] In another study conducted by Zore et al in 2011, it was found that the tumor size did not have any influence on the increased temperature.[18]

One of the limiting factors affecting DA of MG is a higher breast density, but such a limitation was not observed with BT in our study, although this was not statistically significant (p-value of 0.29). We compared breast density with concordance of BT and MG findings. Among the category of “extremely dense fibroglandular parenchyma,” 62.5% cases had a positive correlation (i.e., concordance with MG), but 37.5% cases had a negative correlation. Among the category of “heterogeneously dense fibroglandular parenchyma,” 57.72% cases had positive and 42.28% cases had negative correlation with MG. In the “scattered fibroglandular parenchyma” type, 55.81% had positive and 44.19% cases had negative correlation with MG. In the “predominantly fatty parenchyma” type, 25% cases had positive and 75% cases had negative correlation with MG. Thus, concordance with MG was more for extremely dense breasts than heterogeneously dense and scattered fibroglandular parenchyma and least for fatty parenchyma.

Higher BI-RADS categories reflect higher probability of cancers and hence may have an impact on the extent of thermographic risk stratification. Our study showed that high thermographic risk stratification correlated with a greater number of BI-RADS 5 lesions (31%) as compared with moderate (20%) and low-risk (0%) stratification lesions. All the normal thermographic risk stratification breasts belonged to the BI-RADS 1 category. However, among all high thermographic risk stratification breasts, 33% were BI-RADS 1, 31% were BI-RADS 5, 15% were BI-RADS 4c, 11% were BI-RADS 2, 4% were BI-RADS 4b, 4% were BI-RADS 4a, and 2% were BI-RADS 3; compromising the positive predictive value with no significant statistical correlation with a p-value of 0.275. Similar false positives were also seen by Nathan et al in 1972, showing a false positive rate of 59% and a false negative rate of 29% for BT. Thus, it was concluded that BT had no role in differential diagnosis of symptomatic mammary disease.[19]

A body temperature–dependent modality such as BT is likely to be influenced by physiological and cyclical phases of menstrual cycle, as has also been established in breast MRI; so, we analyzed this stratification in phases. In the first half of menstrual cycle, stroma is dense and cellular with little active secretion[20] and variable vacuolization of epithelial cells.[21] The luteal phase shows stromal edema, basal cell vacuolization, enlargement of ductal lumen, active secretion, and venous congestion. The menstrual phase is characterized by decreased secretion and edema.[20] MRI shows the lowest water content of breast parenchyma during the postmenstrual phase, with a transient increase near the time of menses.[22] [23] [24] Low metabolic activity and less inflammation are seen during the first half of the menstrual cycle.[21] Thus, breast MRI is preferentially performed from day 3 to 14 of the menstrual cycle for optimal signal enhancement and the lowest extraction flow product (ratio of blood volume per weight of tissue per minute [mL/100 g/min])[25]. Similarly, in our study, we divided the menstrual cycle into two phases, Day 3–14 and Day 15–2, and the concordance of BT with MG was compared. In the first half of the menstrual cycle, 42.3% cases had high risk on BT, 14.3% had moderate risk, 14.3% had low risk, and 28.6% were normal on thermographic risk stratification. In the second half of the menstrual cycle, 77.8% had high temperature risk, 11.1% moderate risk, 11.1% low risk, and none were normal. Thus, none of the normal breasts were detected as normal by BT in the second half of the cycle and a greater percentage of cases had high temperature risk stratification. On statistical analysis as per phase-based analysis, nearly similar sensitivity was seen in women imaged in the first half (Day 3–14) and the second half (Day 15–2) of menstrual cycle; however, there was an increase in specificity in the 1st half (28.57%) as compared with the 2nd half (11.11%), with a corresponding increase in DA in the first half (66.67%) as compared with the second half (50%). However, there was no statistically significant correlation, with a p-value of 0.26.

We analyzed if menopausal status had any effect on DA of BT. Our study found concordance of 62% between BT and MG in premenopausal women, 37.5% in peri-menopausal, 53.85% in postmenopausal, and 60% in post-hysterectomy patients. Thus, we found no correlation between the menstrual status of the women and the thermo-mammographic concordance (p-value: 0.49).

Biopsy of lesions were done based on either conventional imaging recommendations or clinical indication. For calculating sensitivity and specificity, histopathology was taken as the gold standard, and conventional imaging as the gold standard if biopsy is not available. The histopathological diagnosis and the thermographic risk stratification correlation show that breasts with malignant lesions had 94% high thermographic risk stratification, 4.8% moderate risk stratification, and 1.2% low risk stratification. However, benign lesions also showed similar high-risk stratification of 84.6% and low-risk stratification of 15.4%. The sensitivity, specificity, and DA were 100.00, 92.59, and 95.96% for conventional imaging. Out of 100 patients included in the study, 31 patients followed up at different time intervals up to 6 months post-BT study. Of these, nine women had normal appearance on conventional imaging but a discordant false positive BT finding. None of these showed new lesions on follow-up MG.

Variable results in different studies have led to variable opinions of BT for screening purpose. In 1972, a large study published by Isard analyzing 10,000 thermograms suggested that BT improved DA when supplemented by MG but was not superior to MG as a standalone technique.[26] In 1977, the National Cancer Institute released guidelines for breast cancer screening, based on the study conducted by the Health Insurance Plan of Greater New York. They concluded that there was no harmful effect of BT, but lack of scientific data supporting its use as a screening technique led to recommendation that BT be discontinued as a screening modality in Breast Cancer Detection and Demonstration Project centers until further research.[27] Feig et al published a study showing that BT was good at detecting larger and malignant lesions, irrespective of breast density. However, in view of high false positive thermograms in the general population, the study was not statistically significant.[8] Some of these publications shifted the focus of breast screening and cancer detection away from BT.[23] [28] In 1982, the United States Food and Drug Administration approved BT as an adjunct to MG to help detect breast cancer.

Improvements in infrared radiation technology and software assistance led to the development of digital infrared thermal imaging (DITI) and computerized thermal imaging. In rodents, increased vascularity, reflecting temperature, was associated with early tumor development, possibly earlier than MG.[28] [29] [30] In 2003, Parisky et al conducted a DITI study on 875 patients, with a sensitivity of 97% and a specificity of 14%, using a dynamic computerized infrared imaging system (BCS2100; Computerized Thermal Imaging, Ogden, Utah), based on pre-cool and post-cool temperatures, in the prone position.[12] Improvement in specificity to 44% with a sensitivity of 97% was seen in 2008 in a similar study by Arora et al on 92 patients (excluding women with large breast sizes) using a different digital thermographic equipment, Sentinel Breast Scan (Infrared Sciences Corp., Bohemia, New York, United States). It was performed in the sitting position in a dedicated equipment suite with a chair equipped with lateral-view side mirrors, an integral air cooler, and a digital infrared camera. Images were obtained after cold stress.[14] In 2010, Wishart et al[31] used the same equipment in 100 patients, prior to core needle biopsy. They analyzed the images in four ways: with DITI, neural network analysis, expert manual review, and using an artificial intelligence (AI) program. The expert manual review was conducted by an independent BT expert blinded to the results but aware of the site of interest. The sensitivity of the expert manual review was 78% and the specificity was 48%. The sensitivity of both routine reporting and neural network analysis was low (53 and 48%, respectively). In 2010, a similar study was conducted by Kontos et al, studying DITI in 63 symptomatic patients. They used the Meditherm med2000 thermal imaging system (Meditherm, Beaufort, North Carolina, United States). In this equipment, the patient was positioned with hands above the head, and three images were obtained from each patient: one face (0 degree) and one oblique on each side at 45 degrees from the middle line. The sensitivity was 25%, specificity was 85%, positive predictive value was 24%, and negative predictive value was 86%. This study concluded that DITI could not be used for breast cancer screening.[32] In 2011, a systematic search of seven biomedical databases by Vreugdenburg et al found a wide variation in sensitivity (0.25–0.97) and specificity (0.12–0.85) of DITI and concluded that there was insufficient evidence to recommend the use of BT for breast cancer screening.[33] The use of computer-aided detection has the potential to decrease the high false positive rate.[34] [35]

Among the recent studies, incorporation of AI into BT has claimed superior results on BT than before, but despite its integration, the results do not appear to be significant to successfully replace the current standard imaging with MG and its supplementary USG assessment.

Martín-Del-Campo-Mena et al used AI to read BT images and used MG as the gold standard. They found the sensitivity of 94.87% and a specificity of 72.26%.[36] Wishart et al,[31] despite using special AI software, did not show encouraging results, with a sensitivity of BT of 72% and a specificity of 48%.

A study conducted on Thermalytix, Niramai, by Singh et al[37] found the sensitivity and specificity of 87.0 and 80.6%, respectively. For women aged ≥45 years, the sensitivity and specificity were 80.5 and 86.5%, respectively. A sensitivity of 80% is considered to be low in women >45 years, with relatively lesser breast density in this age group, in comparison with the performance of newer digital MG images with improved accuracy. The sensitivity of BT even after post-hoc change in the cut-off was still less than that of MG. This study was performed on 258 symptomatic women, which is a small sample. Among these, only 124 (48.1%) women underwent diagnostic mammogram; hence, this makes the sample size further smaller for comparison with MG. In addition, 30 of these women had BI-RADS 0, which were considered as negative analysis on MG. However, routine practice entails a USG assessment to follow a MG for patients with BI-RADS 0. Three of these 30 patients deemed suspicious on subsequent USG were further proven as malignant on histopathology. This endorses the existing MG-USG workflow and may not be taken as negative reading on mammograms. Additionally, BI-RADS 3 findings were considered as positive on MG. Most studies aiming at detecting malignancies will consider BI-RADS 3 as negative, since there is a probability of 98 of 100 such women to be truly negative. This is likely to compromise and explains the poor specificity of MG in this study.

Another study by Bansal et al,[38] using the same device Thermalytix, Niramai, found a sensitivity and a specificity of 99.24 and 88.58%, respectively. Among the reasons for exclusion of patients accrued, a high proportion of 9.3% were excluded due to lack of USG (which was sought to resolve discrepancy between BT and MG results). This is a significant number, since in our study we found a great proportion of false positive uptake on BT in such cases, and excluding such incompletely evaluated cases can skew results. Another 7.1% were excluded due to technical inadequacy, which is a high percentage compared with that encountered for MG in real-world implementation. On analysis of the data, out of 70 patients who were positive on BT, 25 were BI-RADS 0, of which after a correlative/subsequent USG, 3 were deemed as suspicious and were positive for malignancy. Nineteen of all patients were categorized as BI-RADS 4/5, of which 17 were positive for malignancy. This essentially suggests 70 positive BT versus 22 positive MG, highlighting the high number of false positive uptakes on BT in nonmalignant breasts. It may not be appropriate to add BI-RADS 0 to MG analysis without the expected inclusion of USG findings and combined BI-RADS assessment. Interestingly, of the 21 malignancies, 20 were detected by BT, while 19 were symptomatic.

Both studies (Singh et al and Bansal et al) used a similar technique of BT on a similar geographical database. The initial proposed cut-off of 0.5 demonstrated a sensitivity and specificity of 74.6 and 82.1%, and 95.24 and 88.58%, respectively. While Singh et al demonstrated an improved performance with a cut-off of 0.41, this was not utilized in the later study by Bansal et al using despite similar technology.

The recent meta-analysis by Goñi-Arana et al[39] of 22 studies on BT found a pooled sensitivity of 88.5% and a specificity of 71.8%. Of the 22 studies, only 8 studies had a specificity of more than 75%. They found a very high heterogeneity of the sensitivity across the studies with I² value of 79.3%. Heterogeneity for specificity was higher with I² value of 99.1%. The heterogeneity between studies can be attributed to differences in sample size, patient selection criteria, and imaging techniques. Thus, the review concluded that studies showed high sensitivity with variable specificities.

Limitations of the Study

Our study included a small sample size of 100 patients (198 breasts) and was conducted in a diagnostic setup in a tertiary cancer referral center. Hence, BI-RADS 2 lesions were fewer compared with BI-RADS 5 lesions. A larger study is required in a screening setting to assess if it may prove useful as a stand-alone technique. The patients were imaged on the day of MG irrespective of the phase of the menstrual cycle, but this may be an influential criterion in its DA. For calculating the sensitivity of conventional imaging, a follow-up is required. This was a limitation in our study, where a 2-year longitudinal follow-up was not feasible for two patients with BI-RADS 3 lesions on conventional imaging.

Summary Statement

If BT is explored as a screening modality for early detection of breast cancer, its limitations can be a lower detection rate of smaller cancers and false positive uptake in a high proportion of normal breasts, leading to unnecessary anxiety coupled with delayed diagnosis.

Conclusion

BT overall has a high sensitivity, low specificity, low DA, and an unacceptable high false positive rate. Its high thermal uptake also does not reliably differentiate BI-RADS 1 and 5 categories, normal from abnormal, or benign from malignant, and among the malignant lesions, smaller lesions were not as well detected as larger lesions.

Breast density and menopausal status did not have a significant effect on detection or concordance of lesions; however, in premenopausal women, statistically nonsignificant improvement in DA and specificity was seen in BT performed in the first half of the menstrual cycle.

Conflict of Interest

None declared.

• References

• 1 Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2018; 68 (06) 394-424
• 2 Wallis MG, Walsh MT, Lee JR. A review of false negative mammography in a symptomatic population. Clin Radiol 1991; 44 (01) 13-15
• 3 Köşüş N, Köşüş A, Duran M, Simavlı S, Turhan N. Comparison of standard mammography with digital mammography and digital infrared thermal imaging for breast cancer screening. J Turk Ger Gynecol Assoc 2010; 11 (03) 152-157
• 4 Kolb TM, Lichy J, Newhouse JH. Comparison of the performance of screening mammography, physical examination, and breast US and evaluation of factors that influence them: an analysis of 27,825 patient evaluations. Radiology 2002; 225 (01) 165-175
• 5 Hendrick RE. Radiation doses and cancer risks from breast imaging studies. Radiology 2010; 257 (01) 246-253
• 6 Gordon PB. Ultrasound for breast cancer screening and staging. Radiol Clin North Am 2002; 40 (03) 431-441
• 7 Millet I, Pages E, Hoa D. et al. Pearls and pitfalls in breast MRI. Br J Radiol 2012; 85 (1011): 197-207
• 8 Feig SA, Shaber GS, Schwartz GF. et al. Thermography, mammography, and clinical examination in breast cancer screening. Review of 16,000 studies. Radiology 1977; 122 (01) 123-127
• 9 Head JF, Wang F, Lipari CA, Elliott RL. The important role of infrared imaging in breast cancer. IEEE Eng Med Biol Mag 2000; 19 (03) 52-57
• 10 Francis SV, Sasikala M, Bhavani Bharathi G, Jaipurkar SD. Breast cancer detection in rotational thermography images using texture features. Infrared Phys Technol 2014; 67: 490-496
• 11 D'Orsi CJ, Sickles E, Mendelson EB, Morris EA. et al. ACR BI-RADS Atlas, Breast Imaging Reporting and Data System. Reston, Va: American College of Radiology; 2013
• 12 Parisky YR, Sardi A, Hamm R. et al. Efficacy of computerized infrared imaging analysis to evaluate mammographically suspicious lesions. AJR Am J Roentgenol 2003; 180 (01) 263-269
• 13 Omranipour R, Kazemian A, Alipour S. et al. Comparison of the accuracy of thermography and mammography in the detection of breast cancer. Breast Care (Basel) 2016; 11 (04) 260-264
• 14 Arora N, Martins D, Ruggerio D. et al. Effectiveness of a noninvasive digital infrared thermal imaging system in the detection of breast cancer. Am J Surg 2008; 196 (04) 523-526
• 15 Williams KL. Thermography in the prognosis of breast cancer. Bibl Radiol 1969; 5: 62-67
• 16 Faustino-Rocha AI, Silva A, Gabriel J. et al. Ultrasonographic, thermographic and histologic evaluation of MNU-induced mammary tumors in female Sprague-Dawley rats. Biomed Pharmacother 2013; 67 (08) 771-776
• 17 Sterns EE, Zee B, SenGupta S, Saunders FW. Thermography. Its relation to pathologic characteristics, vascularity, proliferation rate, and survival of patients with invasive ductal carcinoma of the breast. Cancer 1996; 77 (07) 1324-1328
• 18 Zore Z, Filipović-Zore I, Stanec M, Batinjan G, Matejčić A. Association of clinical, histopathological and immunohistochemical prognostic factors of invasive breast tumors and thermographic findings. Infrared Phys Technol 2015; 68: 201-205
• 19 Nathan BE, Burn JI, MacErlean DP. Value of mammary thermography in differential diagnosis. BMJ 1972; 2 (5809): 316-317
• 20 Vogel PM, Georgiade NG, Fetter BF, Vogel FS, McCarty Jr KS. The correlation of histologic changes in the human breast with the menstrual cycle. Am J Pathol 1981; 104 (01) 23-34
• 21 Longacre TA, Bartow SA. A correlative morphologic study of human breast and endometrium in the menstrual cycle. Am J Surg Pathol 1986; 10 (06) 382-393
• 22 Cyclic changes in composition and volume of the breast during the menstrual cycle, measured by magnetic resonance imaging - FOWLER - 1990 - BJOG: An International Journal of Obstetrics & Gynaecology - Wiley Online Library [Internet]. [cited 2025 Jul 13]. Accessed October 28, 2025 at: https://obgyn.onlinelibrary.wiley.com/doi/abs/10.1111/j.1471-0528.1990.tb02546.x
• 23 Dean KI, Majurin ML, Komu M. Relaxation times of normal breast tissues. Changes with age and variations during the menstrual cycle. Acta Radiol 1994; 35 (03) 258-261
• 24 Graham SJ, Stanchev PL, Lloyd-Smith JO, Bronskill MJ, Plewes DB. Changes in fibroglandular volume and water content of breast tissue during the menstrual cycle observed by MR imaging at 1.5 T. J Magn Reson Imaging 1995; 5 (06) 695-701
• 25 Delille JP, Slanetz PJ, Yeh ED, Kopans DB, Garrido L. Physiologic changes in breast magnetic resonance imaging during the menstrual cycle: perfusion imaging, signal enhancement, and influence of the T1 relaxation time of breast tissue. Breast J 2005; 11 (04) 236-241
• 26 Isard HJ, Becker W, Shilo R, Ostrum BJ. Breast thermography after four years and 10000 studies. Am J Roentgenol Radium Ther Nucl Med 1972; 115 (04) 811-821
• 27 Response of the National Cancer Institute. NIH/NCI Consensus Development Meeting on breast cancer screening. Yale J Biol Med 1978; 51 (01) 9-11
• 28 Poljak-Blazi M, Kolaric D, Jaganjac M, Zarkovic K, Skala K, Zarkovic N. Specific thermographic changes during Walker 256 carcinoma development: differential infrared imaging of tumour, inflammation and haematoma. Cancer Detect Prev 2009; 32 (5–6): 431-436
• 29 Song C, Appleyard V, Murray K. et al. Thermographic assessment of tumor growth in mouse xenografts. Int J Cancer 2007; 121 (05) 1055-1058
• 30 Xie W, McCahon P, Jakobsen K, Parish C. Evaluation of the ability of digital infrared imaging to detect vascular changes in experimental animal tumours. Int J Cancer 2004; 108 (05) 790-794
• 31 Wishart GC, Campisi M, Boswell M. et al. The accuracy of digital infrared imaging for breast cancer detection in women undergoing breast biopsy. Eur J Surg Oncol 2010; 36 (06) 535-540
• 32 Kontos M, Wilson R, Fentiman I. Digital infrared thermal imaging (DITI) of breast lesions: sensitivity and specificity of detection of primary breast cancers. Clin Radiol 2011; 66 (06) 536-539
• 33 Vreugdenburg TD, Willis CD, Mundy L, Hiller JE. A systematic review of elastography, electrical impedance scanning, and digital infrared thermography for breast cancer screening and diagnosis. Breast Cancer Res Treat 2013; 137 (03) 665-676
• 34 Ng EY, Chen Y, Ung LN. Computerized breast thermography: study of image segmentation and temperature cyclic variations. J Med Eng Technol 2001; 25 (01) 12-16
• 35 Mambou SJ, Maresova P, Krejcar O, Selamat A, Kuca K. Breast cancer detection using infrared thermal imaging and a deep learning model. Sensors (Basel) 2018; 18 (09) 2799
• 36 Martín-Del-Campo-Mena E, Sánchez-Méndez PA, Ruvalcaba-Limon E. et al. Development and validation of an infrared-artificial intelligence software for breast cancer detection. Explor Target Antitumor Ther 2023; 4 (02) 294-306
• 37 Singh A, Bhat V, Sudhakar S. et al. Multicentric study to evaluate the effectiveness of Thermalytix as compared with standard screening modalities in subjects who show possible symptoms of suspected breast cancer. BMJ Open 2021; 11 (10) e052098
• 38 Bansal R, Collison S, Krishnan L. et al. A prospective evaluation of breast thermography enhanced by a novel machine learning technique for screening breast abnormalities in a general population of women presenting to a secondary care hospital. Front Artif Intell 2023; 5: 1050803
• 39 Goñi-Arana A, Pérez-Martín J, Díez FJ. Breast thermography: a systematic review and meta-analysis. Syst Rev 2024; 13 (01) 295

Publication History

Article published online:
10 February 2026