Quentir Medicine Monitor
Evidence-based insights for quantum medicine. Published by Quentir Systems LLC · October 27, 2026.
Five radiologists from the University of Florida College of Medicine in Jacksonville and MD Anderson Cancer Center, writing with Harshada Gothankar of Johns Hopkins University, have produced a guide to quantum machine learning for practicing radiologists. The 15-page primer appeared on arXiv on 23 October 2026, with Mayur Virarkar of UF Health Jacksonville as corresponding author.
The authors start from a problem every imaging department knows. Deep learning models for volumetric MRI and CT need computing power and memory that grow quickly with image size, and large architectures such as ResNet, DenseNet and vision transformers are hard to deploy where hardware is limited. Quantum computing has been proposed as one answer, and the primer asks how far that answer has come in medical imaging.
Nearly all published systems are hybrid quantum-classical pipelines tested on small, downsampled benchmark datasets or on simulators, and the authors write that clinical deployment is not yet feasible. A radiologist reading the paper gets a map of a young research field, with its strongest claims and its weakest tests set side by side.
The paper is a narrative primer. The authors searched PubMed, arXiv and academic search engines through August 2026, prioritized peer-reviewed original studies, and state that they verified all citations against publisher records. They report no specific funding and no relevant conflicts of interest.
What the University of Florida and MD Anderson team explains about quantum hardware
The first half teaches five concepts: qubits, superposition, entanglement, quantum gates and quantum circuits. The authors stress a point that vendor material often skips. An n-qubit register can hold 2n basis states at once, yet a measurement returns a single classical outcome, so any practical speedup depends on algorithm design, data encoding and measurement strategy.
Today's machines belong to the noisy intermediate-scale era, with tens to a few hundred qubits and no full error correction. The primer describes three platforms. Superconducting processors dominate medical imaging research and are reachable through IBM's cloud; trapped-ion systems hold their quantum states longer at smaller scale; photonic chips run at room temperature but remain hard to control and appear rarely in imaging studies.
The authors then explain how images enter a quantum computer at all. Two encodings recur in the literature: the flexible representation of quantum images (FRQI), which stores pixel intensity in probability amplitudes, and the novel enhanced quantum representation (NEQR), which stores grayscale values in basis states and allows direct readout at the cost of extra qubits. Encoding is a bottleneck, because qubit-efficient methods need deep, error-prone circuits.
Quantum pillar: computing. Technology readiness: TRL 3 of 9. Quantum image classifiers have shown proof of concept on small, downsampled public image sets and on simulators, with a few runs on cloud quantum hardware, and the primer's authors write that clinical deployment is not yet feasible.
Which imaging tasks the primer covers: brain MRI, chest radiographs, mammograms and CT reconstruction
Image classification is the most developed application. A four-qubit quantum convolutional neural network sorted 3,264 brain MRI images with 89 percent accuracy for tumor versus no tumor, and only 62 percent when it had to name the tumor subtype. The authors read this as a sign that current models handle simple binary questions better than clinical differential diagnosis.
Chest radiographs supply several of the reported gains. A quantum network working on small PneumoniaMNIST patches reached 83.33 percent validation accuracy against 73.33 percent for a comparable classical network, and replacing one layer of a classical network with a variational quantum circuit raised the area under the curve from 0.727 to 0.836 across 30 training runs on pediatric radiographs. Quantum transfer learning reached an area under the curve of up to 0.93 for cardiomegaly.
Breast imaging shows the more modest picture. Quantum transfer learning reached 84 percent accuracy on full-field digital mammograms, which matched classical transfer learning. On BreastMNIST ultrasound images a hybrid model scored 86.54 percent against 84.17 percent for a classical network. Quentir found a sharper result in September, when simulated quantum kernels lost all 24 matched tests to a default classical support vector machine on brain MRI and breast ultrasound data, which is the kind of baseline comparison the primer asks for.
Reconstruction work is earlier still. One study rebuilt CT images from sinogram patterns as a quantum optimization problem, in simulation only, and hybrid adiabatic quantum computing has been tested for PET and SPECT reconstruction on small simulated images. In radiomics, quantum annealing on a D-Wave system selected features for telling large brain metastases from high-grade gliomas, and the resulting model matched classical classifiers.
Why noise on IBM's 127-qubit Cleveland processor and small test sets limit the results
Hardware noise has a measurable cost. The primer cites a benchmark of medical image classifiers on IBM's 127-qubit Cleveland system in which device noise alone cut accuracy by up to 15.70 percentage points compared with a noise-free simulation of the same model. Gradients in randomly initialized circuits can also vanish as qubit numbers grow, a problem known as barren plateaus, which limits how deep a trainable model can be.
The datasets add a second limit. Small qubit counts force researchers to downsample images, compress features or split circuits, so a model that classifies a 28-by-28-pixel patch says little about a full diagnostic CT volume. The authors flag one reported Dice score of 98 percent for brain tumor segmentation on BraTS 2020 as needing cautious reading because of its nonstandard evaluation.
A larger review reached a similar view. The 2025 systematic review of quantum machine learning for digital health by Gupta and colleagues in npj Digital Medicine, which the primer cites, screened 4,915 studies, kept 169 as eligible and then excluded 123 for insufficient rigor. Only 16 used quantum hardware or noisy simulations, and the reviewers found no consistent trend of quantum models outperforming classical ones in health care. Frederik Flöther's 2023 survey of quantum computing applications in health and medicine in Research Directions: Quantum Technologies supplies the primer's background on reconstruction algorithms.
Advantage also depends on the problem. Shor's and Grover's algorithms show that quantum speedups apply to specific classes of tasks, a point the primer makes early. Quentir described a related mathematical result for sequence data in a post on a proof of an unbounded quantum memory advantage in online classification, where the gain holds for one carefully specified task.
What radiology departments and imaging buyers can watch for in 2027
For patients, nothing changes yet. The primer describes quantum computing in radiology as predominantly experimental, with hybrid proof-of-concept systems limited by current hardware. Its authors write that clinical value stays unproven until studies move beyond simulators, small datasets and weak or inconsistent baselines.
Three kinds of study would show real progress. One is a study that tests a quantum or hybrid model on full-resolution clinical images from more than one hospital, against a strong classical baseline with the same preprocessing. Another is a hardware result that holds its accuracy on a real processor instead of a simulator. A third is reproducible code and data, so other groups can repeat the comparison.
The primer also points to work that needs no quantum advantage at all. Quantum-inspired segmentation runs on ordinary computers, and quantum-based encryption and watermarking schemes aim at images sent between hospitals for teleradiology. The authors close by asking radiologists, physicists and computer scientists to work together so that quantum methods stay tied to real diagnostic problems.
Sources
Primary source: the arXiv primer of 23 October 2026 by Ruben Ortiz Cordero, Gaven K. Smart, Harshada Gothankar, Emilio Supsupin, Priya Bhosale and Mayur Virarkar of the University of Florida College of Medicine in Jacksonville, Johns Hopkins University and MD Anderson Cancer Center. Also drawn on: the 2025 npj Digital Medicine systematic review by Gupta and colleagues, Frederik Flöther's 2023 survey in Research Directions: Quantum Technologies, and earlier Quentir coverage.