PrecisionView is a handheld endomicroscope that overcomes limitations in medical imaging by combining advanced optics with deep learning. Credit: Rice University.
AI Microscope Sees Roughly Five Times More Tissue Than Similar Devices
In A Nutshell
- A pen-sized AI microscope views roughly five times more tissue than similar-detail microscopes.
- It shows cells and blood vessels without refocusing.
- It flagged a lab-confirmed precancerous lesion in a removed cervical specimen.
- Tests in living patients with precancer are still needed.
Catching cancer early can improve a patient’s chances of survival, yet the microscopes built to examine suspicious tissue inside the body can view only a spot smaller than a pinhead at a time. Now, a newly developed handheld device about the size of a pen, powered by artificial intelligence, covers several times more tissue than comparable microscopes without losing the cell-level detail doctors need.
Called PrecisionView, the device was developed by engineers and physicians at Rice University and the University of Texas MD Anderson Cancer Center, both in Houston, and described in the Proceedings of the National Academy of Sciences (PNAS). In early tests, it mapped living tissue inside volunteers’ mouths and revealed a precancerous lesion, later confirmed by lab analysis, in tissue removed from a cervical patient.
Cancers that start in the thin layers of tissue lining organs and body surfaces make up an estimated 80 to 90 percent of all cancer cases. Most patients with these cancers are diagnosed late. Biopsy, the standard way to confirm a diagnosis, means cutting out tissue, and when a suspicious patch of tissue, called a lesion, is large and uneven, only a small portion can be sampled. Because most suspicious lesions turn out to be harmless, even experts struggle to decide where and when to take a sample.
Existing Early Cancer Detection Microscopes Usually See Under Half a Square Millimeter
Microscopes that can examine living tissue cell by cell already exist, but they cover a patch usually under half a square millimeter. Suspicious lesions often span about a square centimeter. Scanning one through such a tiny window is like trying to read a newspaper through a keyhole, slow and full of gaps.
Depth is a second hurdle. Tissue surfaces are bumpy on a microscopic scale, and the details doctors want sit at different levels. Cell nuclei, the command centers of cells, are near the surface, while tiny blood vessels, which also change as precancer develops, lie farther down. Conventional microscopes keep only a thin slice in focus, so doctors must stop and refocus to switch between the two.
AI-Designed Optics Give the Microscope Roughly Five Times the Viewing Area
PrecisionView gets around both problems with a specially shaped optical piece, called a phase mask, tucked inside the microscope. It blurs incoming light in a controlled, consistent way, so the image stays usable across a wide area and a range of depths. An AI system helped design that piece, and a second AI program was trained alongside it to clear up the blur and bring fine details back into focus almost instantly.
Compared with conventional microscopes of similar detail, the result is roughly five times the viewing area and eight times the depth kept in focus. Cell structure shows up as a glow from a dye applied to the tissue, and blood vessels show up in reflected light. Clinicians can watch live video on a screen as the view alternates between the two, with no refocusing. Parts for the device cost about $3,000. It also held up in routine use: over more than a year at multiple locations, including hospital settings, it needed no recalibration or retraining.
Handheld AI Microscope Spotted a Precancerous Lesion in a Removed Cervical Specimen
Researchers first compared the device with a conventional version built from identical parts, minus the AI-designed piece and image-cleanup software. On lab test samples, pig tongue, and postmortem human breast tissue, the conventional version blurred toward the edges and at depths away from its best focus, while PrecisionView stayed consistently sharp.
Next, six healthy adult volunteers had areas inside their mouths dabbed with the dye. Researchers swept the device across the lip and tongue, pausing briefly at each spot, then stitched the frames into maps covering about 1.3 square centimeters. Each session lasted only one to two minutes. Blood vessel patterns differed from region to region, with looping capillaries in some parts of the lip and denser, branching vessels in others.
Cervical tissue provided the most telling test. Nine patients were having tissue removed as treatment for precancer, and each specimen was scanned right after removal. Imaging quality was consistent across all nine, and the paper shows results from two patients.
In the featured specimen, a five-minute scan covered about 2.8 square centimeters and picked out a suspicious patch with crowded cell nuclei, a brighter glow, and more blood vessels than normal tissue. Lab analysis later confirmed a high-grade precancerous lesion. The map also outlined the line where two tissue types meet, a landmark that helps doctors pinpoint high-risk areas of the cervix.
Larger Studies Are Needed Before Clinical Use
Even so, the volunteers were healthy and the cervical scans were done on removed tissue, so the study did not test PrecisionView on precancer inside a living patient. How accurately it separates precancer from harmless tissue was not measured, and the authors call for larger patient studies to find out.
Still, a pen-sized tool that can map microscopic warning signs across large patches of tissue, built from affordable parts, could eventually help doctors decide which patients truly need a biopsy and exactly where to take it, even in clinics that cannot afford bigger machines.
Disclaimer: This article summarizes research for general information and is not medical advice. PrecisionView is an experimental prototype. Consult a doctor with health concerns.
Paper Notes
Limitations
Fluorescence imaging needs an exposure of about 65 milliseconds, which causes visible motion blur when the probe moves across tissue. To avoid it, researchers used a stop-and-go approach, holding the device still for about two seconds per site. The AI’s image-cleanup speed depends on hardware, reaching 22 frames per second on a workstation with a high-end graphics card and 7 on a laptop. Peak sharpness at the exact point of best focus may be modestly lower than in a conventional system, a trade-off of the design that extends focus. The device is built for surface tissue, and deeper or highly varied tissue can scatter light and distort images. Oral imaging involved six healthy volunteers, and cervical imaging was a pilot study of nine patients’ specimens imaged ex vivo, meaning outside the body. How often the device correctly flags or clears tissue was not measured. PrecisionView views tissue parallel to its surface, while lab slides are cut perpendicular to it, which complicates direct comparison. Imaging also required a dye applied to the tissue.
Funding and Disclosures
Support came from the National Institute of Dental and Craniofacial Research (awards R01DE029590 and R01DE032051), the National Science Foundation (grants IIS-1730574 and IIS-1652633), and the National Cancer Institute (grants P30CA016672 and U54CA277843). The Defense Advanced Research Projects Agency also partially sponsored the work through Cooperative Agreement D20AC00002. Author Jacob T. Robinson is cofounder of, holds equity in, and receives payment from Motif Neurotech. The remaining authors declare no competing interests. The article is open access under a Creative Commons Attribution License 4.0 (CC BY).
Publication Details
Authors: Huayu Hou, Jimin Wu, Jinyun Liu, Vivek Boominathan, Argaja Shende, Karthik Goli, Jennifer Carns, Richard A. Schwarz, Ann M. Gillenwater, Preetha Ramalingam, Mila P. Salcedo, Kathleen M. Schmeler, Tomasz S. Tkaczyk, Jacob T. Robinson, Ashok Veeraraghavan, and Rebecca R. Richards-Kortum. Huayu Hou and Jimin Wu contributed equally. Correspondence: Ashok Veeraraghavan and Rebecca R. Richards-Kortum. | Journal: Proceedings of the National Academy of Sciences (PNAS), Volume 123, Issue 20, e2602705123 | Paper Title: “Deep-learning endomicroscope with large field-of-view and depth-of-field for real-time in vivo imaging of epithelial cancer hallmarks” | DOI: https://doi.org/10.1073/pnas.2602705123 | Published: May 11, 2026 (received January 23, 2026; accepted April 9, 2026)







