A 2025 review makes the strongest current scientific case for breast thermography. Read beside the advertising, it describes a different practice — and says so.
Anyone who writes about breast thermography for long enough is eventually handed a paper. Since the summer of 2025 it has usually been this one: a review arguing that radiologists dismissed thermography on the strength of a trial run in the 1970s with equipment and training that would embarrass anybody today, and that the position statements written against it are overdue for revision.
It is a real paper, in an indexed journal, open access, by authors who declare no commercial interest and are funded by a national science ministry. It is not a brochure. It deserves to be read rather than waved away, and this page reads it.
The question it raises is not whether the authors are right about radiologists. It is narrower and more useful than that: what does the review actually endorse — and is that the thing being sold? The answer is no. And the clearest statement of that is in the review itself.
This page takes the review at its word. It does not allege fraud, bias or hidden interest; the authors declare none, the funding is public, and the paper is published under an open licence that permits anyone to check every sentence discussed here.
It does not try to settle whether infrared imaging will one day earn a place in breast care. That is a question for trials that have not been run.
What it does is set the practice the review argues for against the practice advertised to Canadian women, using the review's own text and its own summary table as the source.
No clinic is named on this page. Claims in the right-hand column of Table 2 are described as they appear in Ontario thermography marketing read in 2026, and are documented with dates on the sibling pages of this section. That is the same posture used on What the reading laboratory published.
| Title | Why Do Radiologists Disown Breast Thermography? A Critical Review of Recent Studies and Recommendations |
|---|---|
| Authors | Ane Goñi-Arana, Jorge Pérez-Martín, Francisco Javier Díez |
| Affiliation | Department of Artificial Intelligence, Universidad Nacional de Educación a Distancia (UNED), Madrid |
| Journal | Cancers (MDPI), 2025, volume 17, article 2195 |
| Type | Review — not a clinical study, and not a systematic review |
| Dates | Received 12 May 2025; accepted 27 June 2025; published 29 June 2025 |
| Licence | Open access, Creative Commons Attribution |
| Funding | Spanish Ministry of Science and Innovation, co-financed by the European Regional Development Fund |
| Declared conflicts | None |
| Load-bearing evidence | The authors' own earlier meta-analysis (Systematic Reviews, 2024): pooled sensitivity 88.5%, pooled specificity 71.8%, drawn from the 22 studies that met minimal quality criteria |
Stripped to its recommendations, the paper asks the radiology community for the following:
The review's closing recommendation is not “use it.” It is “stop saying no, and run the trials that would settle it.”
| Point | The 2025 review | The advertising |
|---|---|---|
| Role | An adjunct. The paper repeatedly rules out replacement, and endorses the FDA's warnings against anyone promoting it as a substitute for mammography. | Offered as a radiation-free, painless alternative for women who do not want a mammogram. |
| Interpretation | Computer algorithms analysing the exact temperature of each pixel; in recent work, deep convolutional networks. | Colour images graded by eye on a five-point thermal scale. |
| Who reads the image | Assumed to be a clinical setting with radiological oversight; the whole paper is addressed to radiologists. | Captured by an operator who is not a regulated health professional; graded by a central reader elsewhere. |
| Output | A probability, with the findings it rests on — vascular pattern, temperature difference — stated, so that a clinician can disagree with it. | A grade on a scale, returned with a risk narrative and an invitation to further consultation. |
| A negative result | Must be validated against biopsy or five-year follow-up. The paper notes that counting unverified positives as false positives already understates specificity. | A low grade is returned as reassurance, with a rescreen suggested in a year. |
| Detection claims | Makes no claim of a lead time over mammography, and no claim of a minimum detectable lesion size. | Detection years ahead of mammography, and at a stated tumour size. |
| Evidence cited | 22 studies met minimal quality criteria out of hundreds published; pooled specificity 71.8%, with heterogeneity so extreme the authors report it themselves. | Round numbers — hundreds of supporting studies, sensitivity in the mid-nineties — without the specificity figure that accompanies them. |
| Regulatory status | Notes that the FDA's 1982 clearance was as an adjunct, and that the agency has issued warning letters over standalone promotion. | Sold standalone. |
This is the paper's own summary table, rearranged so that the two columns that matter sit beside the population that produced them.
| Study (year) | Who was imaged | Sens. | Spec. | Note |
|---|---|---|---|---|
| Parisky (2003) | 769 patients scheduled for biopsy | 97% | 14% | The FDA denied clearance for this system in 2002; the manufacturer faced shareholder class actions |
| Arora (2008) | 92 women scheduled for biopsy | 96.7% | 26.5% | The maker appears to have gone out of business |
| Wishart (2010) | 100 patients scheduled for biopsy | 70% / 48% | 48% / 74% | Two systems compared; neither pairing is usable |
| Collet (2014) | 99 patients scheduled for biopsy | 78.8% | 48.6% | Independent evaluation. The authors declined to recommend thermography even as an adjunct. |
| Sella (2013) | 256 healthy women and 178 known cancer patients | 90.9% | 72.5% | Known cases against screened normals — the design most likely to flatter a test |
| Sklair-Levy (2016) | 226 women at high genetic risk | 87.5% | 84.3% | Conference presentation; the sensitivity rests on eight cancers |
| Hellgren (2019) | 1,727 asymptomatic women with dense breasts | 58.3% | 87% | The closest thing in the review to actual screening — and the lowest sensitivity in the table |
| Kakileti (2020) | 470 women, about half symptomatic | 91.0% | 82.4% | Funded by the system's manufacturer |
| Singh (2021) | 258 symptomatic women | 82.5% | 80.5% | Funded by the system's manufacturer |
| Bansal (2023) | 459 women at a secondary care hospital | 95.2% | 88.6% | Funded by the system's manufacturer |
| Gutierrez-Delgado (2010) | 911 women attending screening programmes | 94.1% | not reported | False positives were not counted |
| Rassiwala (2014) | 1,008 asymptomatic women | 97.6% | 99.2% | Only the 49 with abnormal thermograms were worked up — every one of them had a palpable lump. The other 959 were never verified. |
| Yao (2014) | 2,036 women with abnormal mammography or ultrasound | 84.4% | 94.0% | Reported as outperforming mammography on tumours under 2 cm |
| Garduño-Ramón (2017) | 454 volunteers | 86.8% | 89.7% | Recruitment not described |
| Wang (2023) | 2,202 screening patients | 84.5% | 83.9% | Phone-mounted thermal camera |
Sensitivity is the number that appears in advertising. It is also the easy number: a test that calls almost everyone abnormal will catch almost every cancer. That is exactly what the top two rows of the table are — 97% sensitivity at 14% specificity, and 96.7% at 26.5%.
And the high sensitivities come overwhelmingly from women who were already on their way to a biopsy. A cohort assembled because something is wrong is not a screening population, and accuracy measured in the one does not transfer to the other. The review acknowledges this in principle. Its discussion then does not apply it.
Take the review's own pooled figures — 88.5% sensitivity, 71.8% specificity — and apply them to a thousand asymptomatic women, of whom perhaps five have breast cancer.
Roughly four of the five cancers are found. And roughly 280 healthy women are told their result is abnormal. Fewer than two positives in a hundred would be a cancer.
The paper reports this specificity and, in its discussion, calls the pooled figures comparable to those of mammography. Screening mammography's specificity is in the neighbourhood of ninety per cent. The two numbers are not comparable, and the difference between them is the two hundred and eighty women.
I² estimates how much of the variation between studies is real difference rather than chance. The authors report 79.3% for sensitivity and 99.1% for specificity in their meta-analysis, and say plainly that this makes direct comparison of the results difficult.
At 99.1%, pooling has stopped being a summary and become an average of things that are not the same measurement. The single figure of 71.8% has no population it describes.
The review states the heterogeneity in section 2, and quotes the pooled figure as evidence in section 4.
Hellgren and colleagues, published in European Radiology in 2019, imaged 1,727 asymptomatic women with dense breasts. That is precisely the population thermography is marketed to hardest — young enough that mammography may not be routinely offered, dense enough that mammography is known to perform less well.
The result: the thermographic system detected seven of the twelve cancers, a sensitivity of 58.3%. Five cancers that were visible on mammography were missed by the thermogram. The paper reports this accurately, in a bulleted list, and does not return to it.
The second independent evaluation in the review, by Collet and colleagues in 2014, produced 78.8% sensitivity and 48.6% specificity, and concluded that this was not good enough to recommend thermography as a screening modality — not even as an adjunct to mammography. The review quotes that conclusion and moves on.
Of the studies in the table, the two conducted independently of a manufacturer in a clinical setting are the two that reach the same conclusion this section has reached.
This is the part worth reading twice.
The review's discussion identifies the promotion of thermography as a replacement for mammography by alternative-medicine clinics — it names spas and homeopathic centres — as one of the main reasons the technique has no credibility, and it endorses the FDA's repeated warnings to them. The people writing the strongest available defence of breast thermography regard the people selling it as their problem.
On how images are read in clinics that offer thermography as a mammography replacement, the review says they are examined with the naked eye, as was common in the twentieth century, and that this further contributes to the discrediting
of the technique. Every recent study it credits does the opposite: measured temperature, per pixel, by computer.
Visual grading on a colour scale is not a modest version of what the review supports. It is the thing the review blames.
The commercial systems in the review's table have mostly gone. One was denied FDA clearance in 2002 and its maker was sued by its own shareholders. Another maker appears to be out of business. A third abandoned its thermography product and switched to a tactile device; the website for it is gone. The review attributes these failures to the scepticism of the medical community. There is a second available explanation, which is that products are withdrawn when they do not work well enough to sell.
Put the two documents side by side — the advocates' best paper and the clinic's own page — and the striking thing is not disagreement. It is that they are not describing the same procedure.
One is a computer-vision research programme, asking for prospective trials it does not have, a reporting standard that does not exist, and explainable output with a stated probability, and insisting throughout that nothing it proposes should displace a mammogram. The other is a photograph graded by eye and sold at a set price to a woman who has been given to understand that she does not need the mammogram.
When a paper written by enthusiasts has to spend a paragraph distancing itself from the people selling the thing, the selling is not downstream of the science. The science is being cited, not followed.
Goñi-Arana A., Pérez-Martín J., Díez F.J. “Why Do Radiologists Disown Breast Thermography? A Critical Review of Recent Studies and Recommendations.” Cancers 2025, 17, 2195. doi:10.3390/cancers17132195. Published 29 June 2025 under a Creative Commons Attribution licence. Read in full for this page in August 2026.
The meta-analysis on which it rests: Goñi-Arana A., Pérez-Martín J., Díez F.J. “Breast Thermography: A Systematic Review and Meta-Analysis.” Systematic Reviews 2024, 13, 295.
All sensitivity and specificity figures in Table 3 are as reported in the review's own summary table and text. Where this page characterises a study's population or its funding, that too is from the review.
Corrections are welcome. If anything on this page misstates the paper, or if a Canadian clinic does in fact use computer analysis of measured temperature rather than visual grading, write and it will be corrected, with the correction shown.
This page discusses a published scientific review and publicly advertised claims. It makes no allegation of unlawful conduct against any person or business.