When Iakovos Amygdalos, a surgeon at University Hospital RWTH Aachen, proposed using optical coherence tomography to image liver tissue during tumor resection, no one had ever tried it before. The imaging technique—commonly used to examine the optic nerve—uses light waves to generate detailed 3D cross-sectional scans of tissue in seconds, offering a noninvasive peek inside the body's interior. What Amygdalos envisioned was applying this technology to a critical moment in surgery: the point when doctors must determine whether they've removed all malignant tissue.
Liver cancer surgery is a delicate balancing act. Once a tumor is resected, the removed tissue undergoes frozen-section analysis while the patient remains under general anesthesia. Every minute stretches the procedure, tying up operating room staff and increasing the risk of complications. Amygdalos and his team at the German university hospital wondered if OCT could provide answers faster.
Working alongside the Fraunhofer Institute for Production Technology IPT, the researchers acquired 173 OCT scans from 69 patients—88 showing healthy liver tissue and 85 displaying various tumor types. They then applied a machine-learning technique called anomaly detection to the images. Unlike conventional approaches, anomaly detection trains exclusively on examples of normal tissue, learning to recognize what a healthy liver should look like so it can flag deviations. The advantage: this method works well even when malignant samples are scarce.
The results, published in Scientific Reports, demonstrated that the approach achieved a mean accuracy of 81 percent—sufficient to serve as a decision-support tool during surgery. Classification results are ready within seconds. The system recognizes certain tumor types with particularly high reliability: one variant registered 94.3 percent accuracy, another 84.5 percent. A third proved more challenging at 65.9 percent, though even this figure represents meaningful diagnostic information.
Ulrich Krispel, an anomaly detection specialist at Fraunhofer Austria, called the proof of concept convincing. "What is special about this method is that the model is trained exclusively on good examples—that is, scans of normal liver parenchyma," he explained. "The method then reliably detects deviations from this distribution."
For patients, this technology could eventually mean shorter surgeries and less time under anesthesia. For surgical teams, it offers a complementary tool that works alongside traditional histopathological examination. The path forward involves testing the system in actual operating rooms and shrinking the sensor so it fits smoothly into surgical workflows.
Caroline Girmen from Fraunhofer IPT noted that the project has laid essential groundwork. "The next steps will be to test the technology under real operating conditions and to miniaturize the sensor system so that it can seamlessly integrate into the surgical workflow in the long term," she said. For Amygdalos, the potential is clear: this could become a fast and precise tool for characterizing suspicious liver lesions, making the whole procedure more patient-friendly.