A non-invasive system, developed by the Biomedical Optics Group at CIFICEN (CONICET–UNICEN) under the BIONiRS brand, that generates breast metabolic activity maps to help distinguish benign from malignant lesions.
MamoRef is a diffuse optical imaging system: it illuminates the breast with near-infrared laser light and reconstructs metabolic-activity maps from how that light scatters and is absorbed in tissue. This section follows the technical detail published in the group's two papers — Carbone et al., Phys. Med. Biol. 2024 and Carbone et al., Technology in Cancer Research & Treatment 2026 — so that everything shown here can be traced back to the source.
X-rays and near-infrared light give a diagnosis two different, and largely complementary, kinds of information:
| X-rays (mammography) | NIR light (MamoRef) | |
|---|---|---|
| Radiation | Ionizing | Non-ionizing |
| What it shows | Structural information — tissue density | Metabolic information — oxygenation and vascularization |
| Spatial resolution | High | Lower |
| Equipment complexity / cost | Medium to high | Low |
That's why MamoRef is meant as a complement to mammography, not a replacement: it adds a different kind of information, with simpler, more accessible equipment.
Biological tissue is what physics calls a turbid medium: unlike X-rays, NIR light isn't just absorbed as it passes through — it also scatters many times before exiting. Between 600 and 900 nm, the optical window, hemoglobin absorbs relatively little below that range and water absorbs heavily above it, making this the band where light can penetrate several centimeters of tissue and still be detected.
More formally, a turbid medium is described by two coefficients: the absorption coefficient (μa, how much a photon is lost to absorption per centimeter traveled) and the scattering coefficient (μs, how often it bounces off before changing direction). In healthy breast tissue, at 750 nm, typical values are μa ≈ 0.05 cm⁻¹ and a reduced scattering coefficient μs′ ≈ 8.7 cm⁻¹: scattering dominates heavily over absorption, almost 200 to 1, which is exactly what makes a medium turbid and lets light travel several centimeters of tissue before exiting, instead of being absorbed right away (Carbone, PhD thesis, UNICEN, 2017).
Within that window, the molar extinction coefficients of oxygenated (HbO) and deoxygenated (HbR) hemoglobin differ strongly from each other and change with wavelength, crossing at the so-called isosbestic point, near 800 nm. Choosing wavelengths on each side of that point makes it mathematically possible to separate how much light each form of hemoglobin absorbs.
Molar extinction coefficients for HbR (blue) and HbO (red). Green lines mark 660 and 830 nm, MamoRef's lasers on each side of the isosbestic point. Figure 1, Carbone et al. 2024 (Phys. Med. Biol.).
MamoRef schematic. Figure 2, Carbone et al. 2026 (Technology in Cancer Research & Treatment).
The real assembly of the three lasers and the dichroic mirrors that combine their beams into one. Figure 3, Carbone et al. 2024.
The patient lies face-down (prone) on a stretcher with a transparent acrylic (PMMA) window; she rests one breast at a time on the window, flattened only by her own body weight — no additional mechanical compression.
Below the window, inside a light-tight enclosure, sit the optical components:
| Component | Specification |
|---|---|
| Lasers | 3 continuous-wave diode lasers, 660 nm / 785 nm / 830 nm (Coherent Stingray series), combined into a single beam via dichroic mirrors; power capped at ≤5 mW for safety |
| Scanning | 2-axis galvanometer (ThorLabs GVS012/M) steering the combined beam over a ~6×6 to 10×10 position grid (mean of 68 positions across 42 clinical measurements) |
| Detection | High-sensitivity CCD camera (ThorLabs Kiralux CS505MU) with a 12 mm f/1.4 lens, tilted 30° off the window's axis to avoid capturing the laser's specular reflection |
| Scan time | ~2 s of capture per position → 6 to 10 minutes per breast |
| Control | Custom electronics (STM32F103 microcontroller, 16-bit DAC) and a Python/tkinter GUI for selecting the region of interest and operating the device |
Hardware specifications per Carbone et al. 2024 (Phys. Med. Biol.), sections 2.1–2.3.
For each grid position and wavelength, the camera captures a reflectance image. The processing, detailed in both papers, follows these steps:
Before testing it on patients, the reconstruction algorithm was validated with Monte Carlo simulations (using the GPU-accelerated MCX simulator) on 10 mm spherical inclusions emulating a benign fibroadenoma and an invasive carcinoma: the structural similarity (SSIM) between the reconstruction and the known ground truth was above 0.8 in all cases, and two inclusions separated by 20 mm or more were correctly resolved (at 15 mm, they were not). It was also confirmed that a chest wall at 2 cm depth does not distort the recovered relative concentrations.
The exercise was then repeated with physical phantoms (milk diluted with agarose and two inks of different absorbance standing in for HbO/HbR), measured with the real device: the algorithm correctly recovered the relative proportions between the two dyes, confirming that hardware and software work consistently end to end.
MamoRef has, to date, two peer-reviewed clinical studies, both at Hospital Privado de la Comunidad (HPC), Mar del Plata — a high-complexity hospital with a mastology unit accredited since 2018. These are two distinct studies with distinct goals: we keep them separate here, as published, rather than blending their numbers.
Between designing the device and validating it clinically, there's a less visible piece of work that matters just as much: making sure that lying down on the stretcher and going through the study is a comfortable, clear experience. MamoRef went through two prototype generations, and each one was evaluated on that front too, not only on diagnostic accuracy.
The V2 prototype installed at one of the hospitals where it was clinically validated, with the control software that defines the measurement grid.
The first paper describes the device's design and an initial clinical feasibility study: at the time of writing, 42 measurements had been made on volunteers. The paper presents two representative cases in detail:
Case 1: the green circle marks the region of higher uptake. Figure 13, Carbone et al. 2024.
Case 2: no focal region stands out, consistent with a benign lesion. Figure 14, Carbone et al. 2024.
The second paper is a more formal prospective study, carried out between April and November 2023 at HPC, approved by its ethics board (CIREI) and authorized by the Buenos Aires Province Ministry of Health. 21 patients were included, all of whom already required a core biopsy due to suspicious findings on mammography, ultrasound, or MRI (one was excluded from the final analysis because her lesion could not be adequately biopsied). Of all biopsies performed, 7 were cancer (2 of them in situ); the rest were benign lesions (fibroadenomas, fibrosis, periductal mastitis, among others).
Of the MamoRef studies performed, 4 were discarded for artifacts (incomplete breast support on the window, reflections from scratches on the acrylic, patient movement), leaving 17 evaluable sets of maps. Five NIRS-specialized professionals, blinded to each case's diagnosis, scored each image set on a 6-point scale (0 to 5) modeled after BI-RADS, from "normal tissue" to "highly suggestive of malignancy."
Patient 8: invasive ductal carcinoma (NOS), Luminal B subtype. The region of highest ΔHbR sits at x=20–40 mm, y=40–50 mm (hour 7–9), with an extensive region of higher ΔHbO/total hemoglobin around it —consistent with neovascularization around necrotic tissue— and lower SatO₂ in that same area. Figure 3, Carbone et al. 2026.
At the threshold that maximizes Youden's J statistic on the ROC curve (0.6):
Comparing MamoRef's maps against pathology (biopsies) showed close to 80% agreement. Agreement among the 5 evaluators (Fleiss' kappa) was 0.253 — "slight" to "fair": expected for a first clinical experience, with evaluators who had no prior training interpreting this new kind of image.
ROC curve from the prospective study (AUC = 0.77). Figure 7, Carbone et al. 2026.
| Technique | Accuracy | Sensitivity | Specificity | AUC |
|---|---|---|---|---|
| MamoRef | 70% | 62% | 74% | 0.77 |
| Mammography | 71.8% | 34%–60%* | 80%–91%* | 0.73 |
| Ultrasound | 46.9% | 47%–100%* | 19%–100%* | 0.768 |
| MRI | — | 55%–92%* | 70%–88%* | 0.93 |
*Ranges from different studies cited in the literature (see Table 5, Carbone et al. 2026) — the spread across studies for the same technique is itself a sign of how hard it is to pin down definitive reference values. MamoRef's confidence intervals are wide given the still-small sample size.
From lab prototype to hospital validation: here's how MamoRef has progressed.
MamoRef is an optical mammography system developed by BIONiRS. It uses near-infrared (NIR) laser light instead of X-rays to generate breast tissue metabolic activity maps that help classify lesions as benign or malignant.
The Biomedical Optics Group at CIFICEN (CONICET–UNICEN–CICPBA), in Tandil, Argentina, led by Dr. Juan Pomarico (director) and Dr. Nicolás A. Carbone (the project's lead researcher), under the BIONiRS brand.
No. It uses low-power, non-ionizing near-infrared laser light, unlike X-ray mammography.
No. It's designed as a complement: it adds metabolic information to standard imaging techniques, helping reduce unnecessary biopsies and improve lesion follow-up.
It's at the clinical validation stage (TRL 6), with two peer-reviewed published studies (42 measurements in 2024; 21 patients and an AUC of 0.77 in 2026). Regulatory validation with ANMAT and FDA, and multicenter clinical studies, are the project's next steps.
You can write directly to Nicolás Carbone at ncarbone@ifas.exa.unicen.edu.ar, or visit the contact page.
Two peer-reviewed papers and a visual project summary, available to read and download.
MamoRef is made possible by a team of physicists, physicians, and health institutions working together, based at the Biomedical Optics Group at CIFICEN (CONICET–UNICEN), in Tandil.
