No cancer left behind: a testbed and demonstration of concept for photoacoustic tumor bed inspection

When a patient receives a cancer diagnosis, their oncologist will often describe it as one of several possible stages. In stage 0 or I there's usually just a clump of abnormal cells or a local mass: a small tumor confined to its original site. But not necessarily anything beyond that. Then, higher stages correspond to deeper and more complex spread. What we call "metastasis" from the original site. Often into the blood, through the lymphatic system, and then on to other tissues or organs. When a patient is in an earlier stage, all effort and attention usually goes into making sure the cancer does not metastasize. How do they do that? Well there are a few weapons in the standard oncological toolbelt:

1. Treatments like chemotherapy, radiation, targeted therapy, or hormone therapy. They use drugs, energy beams, or molecular interventions to kill cancer cells, shrink tumors, or block the signals tumors need to grow.
2. Immunotherapies that target the immune system with drugs and can help it learn to recognize, attack, and destroy the cancer itself.
3. Resection. The surgical removal of the cancerous mass.

And often doctors will use more than one of these techniques in parallel. Whatever they have to do to stop or slow the spread. But, all these options are tricky, in their own ways. And that third one, resection, is no exception. If a mass is growing on an organ, or on breast tissue, or on any other otherwise-healthy part of your body, the placement of the cut is everything. The surgeon has to figure out how to remove the tumor completely while preserving as much of the healthy tissue as possible.

• Resect too aggressively, and you'll take an unnecessary, and potentially life-altering, amount of tissue, organ, or function with you.
• Resect too conservatively and you'll leave some of the tumor behind. And those remnants may continue to grow and recur and potentially spread.

But choosing the right incision line is harder than it might sound. Tumors do not always have clean, visible borders. They can be irregularly shaped, they can blend into the surrounding tissue, and they can leave behind microscopic disease that the surgeon cannot see or feel in the moment. So it can be nearly impossible for the physician to figure out, with their naked eye, where the perfect cut should go. And not wanting to over-resect, many surgeons default to being a little too conservative instead. The effect of this can be devastating. According to this paper, upwards of a third of breast resections, for example, end up leaving some tumor tissue behind.

The authors of this study think we can do better than that. They're just not sure how, exactly. They've got a hunch that the answer lies in something called photoacoustic imaging. And they think that if you build surgeons a tool out of that technology, it will allow them to solve the issue. To scan through the surgical cavity after resection but prior to closing the patient up, and see if there's any residual cancer or suspicious tissue left behind.

But, as I said, they're just not 100% on any more details than that. They've got an idea, and a theory, they just need a way to flush it out, and iterate on it, and test it. So in this paper they don't build a physical prototype of a clinical scanner, no. They don't build a finished surgical device, or a diagnostic system, or even a simulation of an operating theater. They don't do any of that. They build a testbed: an experimental platform that is set up to let researchers like them prototype, automate, and compare different ways of scanning the tumor bed. It combines a surgical robot, a depth sensor, an electromagnetic tracker, an ultrasound detector, a laser excitation tool, and a 3D visualizer, so that the user can test scanning paths, and probe positions, source-detector geometries, registration strategies, and detection workflows all under repeatable conditions. The goal, again, is not to build a working product, but to lay the experimental foundation on which such a prototype or system could be built and tested in the future. On today's episode we'll walk through why exactly they think that photoacoustic imaging may be the key to intraoperative tumor detection, how their testbed enables experimentation on that idea, and how they demonstrated that the underlying imaging approach is feasible. Let's dive in.

Before we run through the physical components of this system, we need to understand what photoacoustics is, how it works, and why the authors think it's such a game-changer. Photoacoustic imaging is a hybrid technique that starts with light and ends with sound. The idea is that you send short pulses of laser light into tissue. If that light is absorbed by something in the tissue, the absorbed energy causes a tiny, rapid temperature increase. That temperature increase causes the material to expand slightly, and that expansion generates an acoustic pressure wave. In other words, the tissue converts optical absorption into ultrasound. Then an ultrasound detector listens for those pressure waves, and the system uses the timing and strength of the returned signal to infer where the absorbing material is located. And, importantly, different materials absorb light differently. Blood, melanin, nanoparticles, dyes, and contrast agents can all produce different photoacoustic responses depending on the wavelength of light being used. So if you can introduce a contrast agent that preferentially accumulates in tumor tissue, the tumor becomes the stronger optical absorber, while surrounding healthy tissue produces little or no signal at that wavelength. Then, instead of only asking "what does this tissue look like?" you can ask "what is absorbing light here, and where exactly is that absorber in three-dimensional space?"

That is why the authors think it could be so powerful for tumor-bed inspection.

• Pure optical imaging can be useful, but light scatters heavily in tissue, so surface visibility and line-of-sight become major constraints.
• Ultrasound, on the other hand, travels through tissue much more effectively. But, ordinary ultrasound mostly sees structural differences rather than molecular or optical signatures.

Photoacoustics gives you a way to combine the two: optical contrast from the cancer-targeting absorber, with acoustic detection that can reach below the visible surface. In the authors' envisioned workflow, a contrast agent would be taken up prior to surgery. Then, after the main resection is done, a laser would scan the exposed cavity. If there is residual cancer left behind, it should absorb the light, generate a photoacoustic signal, and reveal its location to the surgeon while they can still do something about it. So the promise is not just better imaging in the abstract. The promise is a practical intraoperative tool that can detect remnants, and tell the surgeon whether they need to do more or not.

Sounds great, right? So what's stopping the authors from just prototyping a device? Why go through the whole building-a-testbed thing? Well, it turns out that this kind of intra-cavity photoacoustic inspection has a lot of unresolved design variables. The laser source can be positioned in different places. The acoustic detector can be inside the cavity, outside the body, fixed, tracked, handheld, or robotized. The source and detector can be co-located in one probe, or separated into two devices. The scan path can be dense or sparse, manual or automated, surface-following or geometry-aware. The cavity itself is also confined, deformable, wet, irregular, and hard to access, which means probe placement, optical coupling, acoustic coupling, and tissue motion all become part of the measurement. So you can't jump straight into building a device. If you did, you'd be locking in a design before you know what will work. That's why they're building a testbed instead. It gives them (and other researchers) a way to isolate those choices, repeat them, compare them, and optimize them before trying to turn any one version into a clinical instrument. That is to say, it facilitates systematic prototyping and comparison under controlled, repeatable, surgery-like conditions. Here's how it's set up:

First, there is a da Vinci Research Kit patient-side manipulator (a research version of the robotic arm used in da Vinci surgical systems). This acts as a positioning system. Instead of using it to cut tissue, the authors use it to move the light source over the surface of a simulated tumor bed. A laser fiber is mounted to the robot end-effector with a 3D-printed holder, and a collimator at the fiber tip shapes the output into a small, controlled beam. That gives the system a repeatable way to illuminate one point on the cavity surface at a time. And since the laser is invisible and potentially hazardous, the entire experiment is run inside a safety enclosure. Then there's an internal webcam pointed at the "phantom" (the model of the surgical cavity) so that the operator can monitor the scan without opening the enclosure.

The acoustic side is handled separately. Instead of putting the laser and ultrasound detector into one combined probe, the authors fix an ultrasound probe outside the phantom. The laser excites the suspected residual cancer from inside the cavity, and the ultrasound probe listens for the pressure wave that travels through the phantom. That separation is effectively the embodiment they're trying to study. And this setup means the laser can inspect the cavity without physically touching or deforming the tumor bed, while the ultrasound probe stays outside the cavity and preserves an acoustic detection path. The tradeoff is that the source-detector geometry becomes much more sensitive. If the ultrasound probe is not positioned so that the photoacoustic wave crosses its imaging plane, the system could miss a real absorber.

To know where the cavity is, where the robot is, and where each scan point should go, the bed combines depth imaging and electromagnetic tracking. An RGB-D camera from Intel is mounted above the phantom to capture both color and depth. And the system uses image filtering to identify the retractor, mask the region inside it, threshold the depth image, and extract a point cloud corresponding to the exposed tumor bed. Those points then become the candidate scan locations. An electromagnetic tracker is then used to follow the retractor and other instruments, so that the system can relate the phantom, the robot, and the imaging data to one shared spatial frame. This is what lets a point detected in the camera view become a physical target that the robot can move the laser toward.

The software stack is built around something called "3D Slicer", a medical imaging platform that acts as the central visualization, registration, acquisition, and navigation environment. Depth data from the camera is streamed into Slicer and converted into the point cloud. Electromagnetic tracking data is streamed in through the "PLUS Toolkit". Communication with the Research Kit is handled through ROS2 and SlicerROS2, which lets the system send motion commands and read position data inside the same navigation environment. Then raw ultrasound channel data is streamed from the acquisition hardware into Slicer through OpenIGTLink. So, at a high level, Slicer is the hub where the cavity surface, scan path, robot pose, tracker pose, ultrasound data, and detected photoacoustic signal are visualized together in 3D.

And once the authors assembled all the pieces, they just needed to figure out a way to test it. Not to build a device on it, but just to demonstrate that the basic inspection concept was actually plausible. That you could, at least in theory, scan a tumor bed point by point, trigger a detectable photoacoustic response from cancer-like material, and map the detected signal back onto the cavity surface. For that, they needed a controllable tumor-bed model with known ground truth. Here's how it worked: first they built breast-tissue-mimicking phantoms out of plastisol and cellulose, carved surgical-cavity openings into them, and painted simulated residual lesions onto the cavity surface using ink and barium sulfate. The ink served as the optical absorber. The barium sulfate made those same regions visible on CT. Then they sealed the lesion material under a plastisol layer, scanned each phantom with CT for ground truth, used the onboard camera to segment the cavity and generate the scan path, registered the point cloud and robot coordinate systems, and commanded the robot to move the laser across the tumor bed. At each scan point, they recorded ultrasound data and marked the point as positive if a photoacoustic signal was detected. Then they compared those detected points against the ground truth visible on the CT.

In the end, the results were promising. The system was able to detect the (simulated) residual cancer in all of the phantoms, and the cavity-sensing and registration pipeline were able to turn an irregular tumor-bed surface into a robot-executable scan path. That being said there were misses and false detections, for sure, but they were arguably more informative than fatal. Because they exposed exactly the kinds of design variables the testbed is meant to study: ultrasound probe placement, source-detector geometry, scan density, tracking error, and the limits of using a fixed external detector. So while the paper does not show that this technology is ready for the operating room, it does show that the basic workflow is experimentally plausible. And that their testbed is useful for discovering what has to be optimized next.

Want to go deeper? Make sure you download the paper. The authors include the full coordinate-frame setup, point-cloud segmentation workflow, robot registration steps, ultrasound calibration process, raw channel-data acquisition pipeline, phantom construction details, quadrant-localization metric, failure-mode analysis, and more.