Active Thermal Imaging Spots Residual Breast Cancer During Surgery in 90 Seconds

Surgeons removing a breast tumor often face an uncomfortable truth at the operating table: they cannot reliably tell by eye or by touch whether every cancer cell has been cleared. A new open-access study in Discover Imaging reports a prototype system that addresses this gap with something surprisingly simple in principle—heat. By briefly warming the surgical field with infrared lamps and filming how the tissue cools, the system distinguished margin-negative from margin-positive resections with a reported classification accuracy of 93.33%, sensitivity of 92.00%, and specificity of 94.44% during intraoperative breast cancer surgery.

The technique, called active thermal imaging, rests on a measurable biological contrast. Cancer tissue has an elevated metabolic rate, higher water content, and greater tissue density than benign tissue, and these properties change how it stores and dissipates heat. The researchers applied controlled infrared radiation to a region of interest, then monitored the transient thermal response continuously with a high-resolution infrared camera. Malignant areas, the study found, cool more slowly in later intervals after heating stops, and that difference in cooling behavior becomes a quantitative signature of residual disease.

The hardware is deliberately straightforward. The prototype combines a FLIR A700 thermal imaging camera with a 640 × 480 pixel microbolometer detector, a spectral range of 7.5–14 micrometers, thermal sensitivity of 0.04 K, and real-time imaging at roughly 30 frames per second, together with four 250-watt infrared heating lamps and a computer for data acquisition and analysis. During calibration, the camera’s temperature readings were compared against K-type thermocouple probes placed on a resected specimen; across probe locations that differed by about 10 K, the camera matched the thermocouple data with a largest deviation of less than 2 K at an emissivity correction setting of 0.95.

Safety was engineered into the protocol from the start. During calibration the tissue surface was deliberately heated above 46 °C to create measurable gradients, but in all subsequent experiments an automatic cutoff kept tissue temperature below 42 °C. The region of interest—typically the deep resection margin, the margin adjacent to the pectoralis major fascia and chest wall that most often emerges positive in breast cancer surgery—was heated from a distance of 20 centimeters while the camera recorded from 65 centimeters. After heating ended, the tissue cooled for roughly 60 seconds under natural convection while the camera captured video at 30 frames per second, each frame a 640 × 480 matrix of instantaneous temperatures.

The data transformation is where the physics becomes diagnosis. The recorded cooling phase was subdivided into temporal subzones, and the rate of temperature change was calculated at every pixel for each subzone. A clustering algorithm condensed these cooling rates, and differences between the first and second subzones and between the first and third subzones were derived as matrices the authors call R12 and R13. Pixels with positive scores indicate slower cooling in later subzones—the pattern characteristic of cancerous tissue and suggestive of a margin-positive location—while negative scores indicate the faster cooling typical of benign tissue. Contour plots of these matrices were displayed on a monitor to give the operating surgeon visual guidance, and the entire procedure from heating to displayed result completed within 90 seconds, which the authors deem real-time in a surgical context.

To validate the approach, the team first ran a cross-sectional analytical study on 26 freshly resected breast cancer specimens at a tertiary cancer hospital, orienting each sample with the deep resection margin facing upward and correlating thermal outputs with histopathological reports. Statistical analysis using logistic regression identified two derived parameters—the normalized scores v(si) of R12 and v(si) of R13—as significant predictors of margin status, each with a p value below 0.001, while the simple mean values of the two matrices failed to reach significance. For the specimens, a decision tree classifier built on these two variables achieved a test accuracy of 95.12%, with sensitivity of 94.12% and specificity of 96.15%, and the logistic regression model yielded an area under the ROC curve of 0.948.

The decisive test came in the operating room. After operating parameters were finalized and safety standards met, the system was deployed during ten breast cancer surgeries—modified radical mastectomies—in willing patients who provided written informed consent under approval from the Institutional Ethics Committee of College of Medicine & Sagore Dutta Hospital. Following complete breast tissue resection, the chest bed was actively thermally imaged, and the normalized scores from each grid and sub-grid were correlated with postoperative histopathological findings. Cutoff values of 68.42 for v(si) of R12 and 71.43 for v(si) of R13 were established for intraoperative decision-making, and the optimized decision tree using these thresholds reached 93.33% accuracy, 92.00% sensitivity, and 94.44% specificity, with an AUC of 0.956 for the refined logistic regression model on chest bed data.

The clinical safety record in this pilot was clean. Monitoring covered intraoperative vital sign stability—heart rate, oxygen saturation, temperature, and blood pressure—as well as any events requiring medical intervention or prolonging surgery, and postoperative checks for skin irritation, burns, inflammation, discomfort, or device-related surgical site infection. No intraoperative or postoperative adverse events were reported across all patients evaluated. Histopathology also confirmed no evidence of tissue necrosis attributable to the infrared heating, and the authors note that analyzing both cluster differences simultaneously helps mitigate false positives caused by surface artifacts such as blood or fluid.

Important caveats temper the enthusiasm, and the authors state them plainly. Although the workflow completes within 90 seconds, all results in this study were generated through post-processing; the system has not yet presented results to the surgeon during live surgery, and prospective clinical validation in live surgical settings is planned. The clinical validation to date is limited to breast cancer within a controlled surgical environment, and broader validation across different solid cancers and more diverse patient populations is needed before generalizability can be claimed. The authors also caution that changes in hardware configuration or imaging geometry can shift absolute temperature readings, emissivity correction, and spatial resolution, meaning model retraining and recalibration of the cutoff thresholds may be required in new settings.

Even with those limits, the contrast with existing intraoperative options is striking. Near-infrared fluorescence methods often rely on exogenous agents, intraoperative MRI and CT are logistically and sterility-challenging in standard operating rooms, and palpation and visual inspection remain the everyday fallback despite their documented insufficiency. Active thermal imaging is noncontact, noninvasive, label-free, and contrast-free, and because margins other than the deep resection margin can be assessed by reorienting the camera and heating source, the authors argue the system could be adapted to any clinically suspected margin. With multicenter trial evaluations proposed as the next step, the study positions dynamic thermal response profiling as a candidate tool for helping surgeons achieve complete tumor clearance on the table—potentially reducing the margin-positive resections that drive recurrence, mortality, and morbidity in breast cancer—rather than waiting days for the pathology report to reveal what the heat already showed.

Subject of Research: Active thermal imaging for real-time intraoperative detection of residual breast cancer

Article Title: Real time intraoperative detection of residual breast cancer using active thermal imaging

Article References: Saha, R., Sarkar, S., Dutta, A., Majumder, R., Datta, A., Ghosh, A., Ganguly, R., Santra, A. K., & Garain, U. (2026). Real time intraoperative detection of residual breast cancer using active thermal imaging. Discover Imaging, 3(1), Article 1. https://doi.org/10.1007/s44352-025-00022-1

Image Credits: AI Generated

DOI: 10.1007/s44352-025-00022-1

Keywords: active thermal imaging, breast cancer, surgical margins, intraoperative imaging, infrared thermography, modified radical mastectomy, decision tree model, histopathology, R0 resection, medical imaging, pilot study, cancer surgery