The first line of treatment for cancer is, whenever possible, to remove the cancerous tissue from the body. Though often remarkably effective, removing only the cancerous tissue is a challenge for doctors and surgeons. With no intraoperative method to analyze excised tissues, a surgeon typically must rely on preoperative visualizations-ultrasounds, MRIs, and the like-to accurately locate cancerous tissue and then postoperative examinations of the excised tissue to determine whether the cancer has been entirely removed.
The challenge for surgeons, of course, is balancing the need to remove all of the diseased tissue while preserving as much as possible of the affected region. For example, lumpectomies are a conservative treatment for breast cancer, and studies have shown that patients survive breast cancer equally well with lumpectomy as with removal of the entire breast. "But the current way to ensure that all the cancer has been removed is through postoperative pathology, and if pathology shows that there are cancerous cells on the boundary of the tissue that's been removed, it's necessary to go back for a second or even a third surgery to remove the rest of the cancer," explains Lihong Wang, Caltech's Bren Professor of Medical Engineering and Electrical Engineering, Andrew and Peggy Cherng Medical Engineering Leadership Chair, and executive officer for medical engineering. "Up to one-third or more of breast cancer patients have to undergo these repeat surgeries."
Answering a challenge from collaborating cancer specialists at the City of Hope in Duarte, California, Wang has pioneered a means of analyzing excised tissues during surgery itself with the help of AI, permitting surgeons to continue removing tissue as necessary until all the cancer is gone.
Traditional imaging of tumor samples requires several steps. The first step is to stabilize the tissue prior to slicing it so that it can be visualized properly on a glass slide under a microscope. One way to "fix" the sample in place is to freeze it, but, as Wang explains, this can "cause all sorts of problems. When you chip away ice, you may tear apart the tissue, and it loses its integrity. And breast tissue, which is fatty, does not freeze well." Another option is to fix the tissue in a chemical called formalin and then embed it in paraffin, but this takes time and can also cause distortions.
After this, the tissue can be sliced into cross-sections. Then the samples are stained with two chemicals, hematoxylin and eosin (H&E), one of which stains cell nuclei and the other of which stains the cytoplasm that surrounds cell nuclei. Visualizing these components makes it possible to see the difference between cancerous cells and adjacent healthy cells; the cancerous cells are more densely packed together than healthy cells, and their nuclei occupy a bigger fraction of the entire cell.
Even when the samples are carefully prepared for viewing, the accuracy of the pathological analysis will depend on the skill of the individual pathologist who examines them. This makes it difficult to standardize histology results. Worst of all, this preparation takes time, causing patients to wait hours to days for pathology results.
Wang's new technique, known as ultraviolet photoacoustic microscopy (UV-PAM), eliminates the need for freezing, fixing, slicing, staining, and even direct examination by pathologists.
Using UV-PAM, tissue is removed from the body and then excited with a low-energy laser. The frequency of the laser is set to the so-called absorption peak of the nucleic acids that make up DNA and RNA (the absorption peak represents the frequency at which the material maximally absorbs electromagnetic radiation), yielding what Wang calls "Mother Nature's natural staining process." Cell nuclei appear brighter than surrounding tissue as they absorb the laser's light.
The absorption also causes the tissue to vibrate, producing ultrasonic sound waves. These sound waves allow for very precise imaging of the sample; the current resolution is between 200 and 300 nanometers. The image is then adjusted through artificial intelligence (AI) to appear as it would with traditional H&E staining so that pathologists and surgeons accustomed to reading standard pathology slides can read these images easily without any further training.
Having been provided with an enormous database of images of various types of tissue samples, the computer can compare its results and provide an initial diagnosis. "This is where AI can shine," Wang says, "because AI can examine the images as quickly as we acquire them. We can simultaneously analyze one area of the tumor while scanning the next one, which speeds up the process even more."
To allow this imaging technique to be useful in the operating theater, surgeons requested an analysis time limit of 10 minutes. "We're confident we can image everything within 10 minutes, if not 5 minutes or shorter-fast enough to guide decisions before the surgeon closes the incision-and we can provide a lot more data than any single pathologist could read," Wang says. Importantly, the technique so far appears to be "tissue agnostic," meaning that it works equally well on breast, bone, skin, and organ tissues.
"We're still in the testing stage with this technology, but we hope to move toward a commercial product that can be widely used," Wang says.
This technology is reported in an article titled " Rapid cancer diagnosis using deep-learning-powered label-free subcellular-resolution photoacoustic histology " and published in Science Advances on November 21, 2025. Co-authors include Wang, Yilin Luo, Cindy Liu, and Samuel Davis of the Caltech Optical Imaging Laboratory; Byulee Park, Rui Cao, and Yide Zhang, formerly of Caltech and now at Pohang University of Science and Technology in Korea, Case Western Reserve University, and University of Colorado Boulder, respectively; Yushun Zeng and Qifa Zhou of the Viterbi School of Engineering at USC; and Massimo D'Apuzzo at the department of pathology at City of Hope Beckman Research Institute and Medical Center in Duarte.
Funding for this research was provided by the National Institutes of Health and the National Research Foundation of the Korean government.
