University of Florida and MD Anderson Radiologists Review Quantum Machine Learning for MRI, Chest X-Ray and CT: What the arXiv Primer of 23 October 2026 Finds
What radiologists at UF Jacksonville and MD Anderson found about quantum machine learning in imaging
Quentir Medicine Monitor
Evidence-based insights for quantum medicine.
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.
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.
Kipu Quantum Ran a Molecular Toxicity Classifier on QuEra's 256-Qubit Aquila: What 200 Neutral Atoms Added for 122 Drug-Candidate Molecules
Kipu Quantum tests neutral-atom features for drug toxicity prediction
Quentir Medicine Monitor
Evidence-based insights for quantum medicine.
The German quantum software company Kipu Quantum has reported a molecular toxicity experiment run on QuEra Computing's Aquila, a neutral-atom quantum processor with up to 256 atoms. Each of 122 molecules from a public toxicity dataset was encoded into the controlled quantum-state evolution of 200 atoms, and the measured patterns were turned into about a thousand new numerical features per molecule for an ordinary machine learning classifier.
On those 122 molecules, drawn from a dataset of 171 laboratory-tested compounds, the quantum features raised the balanced accuracy of the toxic or non-toxic call from about 0.55 to about 0.70, according to the authors. For anyone who screens drug candidates for safety, the result is an early test of whether a quantum device can make a classical prediction model better at spotting the harmful minority of compounds.
The work is described in a technical blog post on the company's site, Reading Toxicity of Molecules with Neutral-Atom Quantum Processors, by Antonio Ferrer Sánchez, Carlos Flores and Alejandro Gómez Cadavid, carrying a publication date of 10 July 2026. Quantum Computing Report brought it to wider attention on 17 October 2026. The post is not a peer-reviewed paper, and it neither links nor cites a preprint with the full methods. The figures below are the authors' own, read from the post.
Aquila is an analog machine. It holds single atoms in place with focused laser beams and drives them with smooth light pulses into a highly excited Rydberg state; when one atom is excited, its close neighbors respond, and this interaction shapes how the whole array evolves. QuEra's product page states that Aquila has been publicly available through Amazon Braket since 2022, with a minimum atom spacing of 4 micrometers.
Kipu Quantum's method, which it calls quantum feature extraction, writes the description of each molecule into the settings of that evolution. The team assigned each molecule its own set of driving schedules and atom spacings, laid out 200 atoms on a grid of 14 by 15 positions a few micrometers apart, and let the system evolve for a little under four microseconds before reading out every atom. The authors write that they reached the machine through their own Kipu Quantum Hub platform.
They submitted the training set one molecule at a time, with 500 independent measurements per molecule, and processed 122 molecules in total, roughly two thirds labeled non-toxic and one third toxic. From the strings of excited and unexcited atoms they built about 1,000 features per molecule. A conventional classifier then made the final toxic or non-toxic decision. The quantum processor never classified anything itself; it acted as a transformation step in front of standard software.
SpinQ and DIVBIO Add a Five-Qubit Circuit to GLP-1 Receptor Potency Prediction: What the arXiv Study of 9 October 2026 Shows
A five-qubit circuit sharpens GLP-1 receptor potency prediction
Quentir Medicine Monitor
Evidence-based insights for quantum medicine.
Researchers at the Shenzhen quantum computing company SpinQ Technology and the Shenzhen drug developer DIVBIO Pharmaceutical report that a small simulated quantum circuit, bolted onto a conventional neural network, makes slightly better predictions of how strongly candidate molecules switch on the GLP-1 receptor. That receptor is the target of the drug class now used widely for type 2 diabetes and weight management, so even modest gains in early compound triage attract attention.
The preprint, posted to arXiv on 9 October 2026 by Chenyan Jia, Yaping Ma, Yuen Wang and Cong Guo, tests what the authors call a quantum-circuit residual head: a five-qubit circuit that adds a correction on top of an ordinary linear prediction. Its subject is potency prediction, the task of estimating from a molecule's structure the concentration at which it produces half of its maximal effect in a laboratory assay. The study also includes a careful classical control, and that control shapes how the result should be read.
The full paper, its 19 pages of supporting information and the abstract are open on arXiv under identifier 2610.13037. It has not yet been peer reviewed.
The team drew its measurements from release 37 of ChEMBL, the open bioactivity database run by EMBL's European Bioinformatics Institute. They kept only records for the human receptor that measured cAMP, the signalling molecule a cell produces when the GLP-1 receptor is activated, and that reported an exact EC50 value. After removing flagged and duplicate entries and merging repeat measurements of the same structure by their median, the GLP-1 receptor set held 1,052 compounds, of which 472 were small molecules and 580 were peptides under the authors' grouping rule, which counts amide bonds.
Two further receptors served as checks. The beta-2 adrenergic receptor (ADRB2) set held 829 compounds, 711 of them small molecules, and the melanocortin-4 receptor (MC4R) set held 767. Both receptors are established drug targets, the first in asthma and the second in rare genetic forms of obesity.
The authors point out a limitation of public potency data. A single cAMP potency value merges measurements taken in different laboratories under different assay conditions, so part of the error any model shows reflects noise in the labels themselves.
Researchers in Taiwan, Korea and at Brookhaven Test a Quantum Attention Model on Merck's Alzheimer's Drug Verubecestat: arXiv 2610.04588, 3 October 2026
A quantum attention model tested on an Alzheimer's drug series
Quentir Medicine Monitor
Evidence-based insights for quantum medicine.
Merck stopped its large Alzheimer's trial of verubecestat early, a pill designed to block the enzyme BACE1, which helps produce the amyloid found in the brains of patients. The drug had been carefully refined molecule by molecule, yet in a trial involving 1,958 participants with mild-to-moderate disease it did not reduce cognitive or functional decline compared with placebo.
That chemistry now has a second life as a test set. On 3 October 2026 five researchers from National Yang Ming Chiao Tung University, KAIST, Taiwan's National Center for High-Performance Computing, Brookhaven National Laboratory and Taipei Medical University posted a study of quantum attention for drug discovery. They used the compound series that led to verubecestat as a case study, asking whether a model with a small quantum circuit inside it orders those molecules by potency the way the laboratory did. The same models were also scored on standard benchmarks, including whether a molecule can cross the blood-brain barrier.
The paper, "Variational Quantum Attention for Molecular Graph Learning," by Yu-Cheng Lin, Yu-Chao Hsu, Tai-Yue Li, Nan-Yow Chen and Samuel Yen-Chi Chen, is a preprint on arXiv (2610.04588) and has not yet been peer reviewed. For a pharmacologist, a hospital formulary lead or a research director weighing quantum computing claims in drug discovery, its value lies in how plainly it reports what changed and what did not.
Software that predicts molecular properties is already a routine filter in early drug discovery. Chemists draw large numbers of candidate structures, and models rank them for solubility, brain penetration or binding strength before anyone spends money on synthesis and testing. A poor ranking sends chemists after the wrong molecules. The question this paper asks is narrow and useful: what happens to such a model when one small part of it is replaced by a quantum circuit?
Florence and Barcelona Researchers Run a Quantum Diffusion Model's Noise Step on IBM's 133-Qubit Torino for Blood-Cell, Brain MRI and Rib CT Images: What arXiv 2609.31070 Shows
Quentir Medicine Monitor
Evidence-based insights for quantum medicine.
Generative AI can now draw convincing medical images: blood-cell smears, brain scans, slices of a CT volume. Hospitals and research groups want such synthetic images to enlarge small training sets, to share data without exposing patients, and to test diagnostic software on rare cases. The most successful generators today are diffusion models, which learn to turn random noise back into a plausible image. A preprint posted on 25 September 2026 asks whether a real quantum computer can take over one half of that process.
The paper, "Quantum Diffusion Models for Medical Image Analysis" (arXiv 2609.31070), comes from Francesco Aldo Venturelli and Miguel A. González Ballester of Universitat Pompeu Fabra and the Barcelona Supercomputing Center, Stefano Martina, Marco Parigi and Filippo Caruso of the University of Florence, and Alba Cervera-Lierta of the Barcelona Supercomputing Center. They built a hybrid quantum diffusion model in which the step that adds noise to an image runs on IBM's 133-qubit ibm_torino processor, while a classical neural network learns to remove that noise again. They then used it to generate synthetic medical images from three public collections: blood-cell photographs, brain MRI slices and 3D rib CT volumes.
The authors report that their hybrid model is competitive with a classical diffusion model of the same design, and better on some measures, especially for the 3D rib volumes. The comparison is limited: 100 generated samples per dataset, small images, and a classical baseline the team reproduced itself. No radiologist judged the images, and no diagnostic task was run on them.
That leaves a narrow but real result. It is one of the few medical-imaging studies in quantum machine learning that runs part of its pipeline on physical quantum hardware with public medical-image datasets, and it deserves a precise reading of what the hardware did and what the numbers show.
Simulated Quantum Kernels Lost All 24 Matched Tests to a Default Classical SVM on Brain MRI and Breast Ultrasound: Journal of Imaging Informatics in Medicine, 21 September 2026
Quentir Medicine Monitor
Evidence-based insights for quantum medicine.
On 21 September 2026 the Journal of Imaging Informatics in Medicine published a study by four computer scientists in Lahore, Abu Dhabi and Manchester that put two quantum machine-learning classifiers against ordinary statistical tools on brain MRI scans and breast ultrasound images. In all 24 matched comparisons the top-scoring quantum model scored lower than the top-scoring classical model, and simulating the quantum models on ordinary computers took roughly 50 to more than 220 times as long as running the classical one.
The paper, by M. Usman Hashmi, M. Adnan Hashmi, Muazzam Ali and Raheem Sarwar, asks whether quantum kernels improve medical image classification once every model gets the same features, the same training budget and the same protection against data leakage. Its answer, for the settings it tested, is no: a standard RBF-SVM with its default settings outperformed both quantum kernels in every condition. The authors describe their design as a leakage-controlled benchmark, and that design is the reason the study deserves a close reading.
Many earlier papers on quantum classifiers for radiology and pathology reported encouraging accuracy figures. Several compared a tuned quantum model with a weak classical one, or tested on a single split of the data, where a small overlap between training and test images can inflate a score. This study sets out to give both sides the same conditions and then measure what is left.
For a radiology department or a hospital buyer, the result gives a practical reference point. When a vendor or research partner proposes a quantum classifier for imaging, the first question is whether it was compared with a strong, well-configured classical baseline under the same conditions. This paper shows how large the difference can be when that comparison is made carefully. It also lists the controls such a comparison needs: identical inputs, preprocessing fitted on training images only, several random seeds and reported calibration.
Zhao, Preskill and Huang's 8 April 2026 Paper Puts a 68,000-Cell Blood Dataset Benchmark Under 60 Logical Qubits: What a Quantum Memory Advantage Means for Hospital Genomics
Quentir Medicine Monitor
Evidence-based insights for quantum medicine.
On 8 April 2026 seven authors from Caltech, MIT, Google Quantum AI and the startup Oratomic, among them John Preskill and Hsin-Yuan Huang, posted a 144-page proof that, for a specified family of classification and dimension-reduction tasks under stated assumptions, a quantum computer of polylogarithmic size reaches a prediction performance that any classical machine would need exponentially more memory to match. The claim reached a wider readership on 11 September 2026, when WIRED Japan explained it to museum visitors as an advantage in size, not in speed.
The paper's medical test case is a genomics one. In a numerical benchmark on the expression profiles of 68,000 peripheral blood cells, the authors report that their method separates cell types and finds the main axis of variation with a calculated requirement of fewer than 60 logical qubits, where the classical methods chosen for comparison need four to six orders of magnitude more memory units. The proposed protocol draws random samples, processes each once and discards it; the benchmark reports performance and an estimated memory requirement, and it did not run that stream on a quantum machine. The advantage the authors prove is one of memory, not speed, and that distinction decides what a hospital genomics group should make of it.
Florida Atlantic's 90.26% Quantum Heart-Disease Result Came From a Simulator: the AI Paper of 21 May 2026 and the FAU Release of 27 August
Quentir Medicine Monitor
Evidence-based insights for quantum medicine.
Florida Atlantic University announced on 27 August 2026 that its engineers had built a quantum machine learning framework for heart disease prediction reaching more than 90 percent accuracy. The figure is precise and it is checkable. It is 90.26 percent, produced by a quantum support vector machine with angle encoding, averaged across five folds of a clinical file holding 918 patients.
The work behind that announcement appeared three months earlier. Muhammad Minoar Hossain, Md. Hasibul Hassan Himal and Arslan Munir published "A Comparative Study of Quantum Feature Maps and Quantum Classifiers for Heart Disease Prediction" in the MDPI journal AI on 21 May 2026, as article 180 of volume 7, under a Creative Commons license that lets anyone read the whole methods section. Section 2.6.2 of that paper states that every quantum experiment ran in a simulation-based environment rather than on a physical quantum processing unit. The university announcement does not carry that sentence, and neither does the coverage that followed it.
That difference decides what the study is evidence for. Simulated qubits establish whether an algorithm has promise in principle; a run on a physical processor establishes whether the machines that exist can deliver it, once noise, limited connectivity and readout error have had their say. Everything else in the paper holds up well under checking. Several of its numbers are more informative than the ones the announcement chose to lead with, and one of them disagrees with the paper's own abstract. The study is a careful piece of comparative work whose careful parts were the first thing lost in transmission.
Which Blood Signal Rewards a Quantum Kernel
A screening gap a blood draw might close
Annual low-dose CT screening lowers lung cancer mortality, and most of the people who qualify for it are not up to date with it. The American Cancer Society reported in June 2024 that 18.1 percent of screening-eligible US adults were up to date in 2022 survey data, which leaves the larger share of a proven survival benefit unclaimed. Blood-based tests are attractive because collection is easier to distribute and to repeat than a CT appointment. The study compares two molecular readouts, drawn from a wider field that also includes mutation panels, circulating proteins and cell-free RNA: cell-free DNA fragmentomics, which reads the length and position of circulating fragments, and methylation, which reads chemical marks at defined genomic targets. Both produce high-dimensional, nonlinear data, and both are degraded by the heterogeneity of lung cancer.
What the Cleveland Clinic and IBM Quantum preprint reports
A team from Cleveland Clinic Research and IBM Quantum posted a preprint on August 19, 2026 asking whether a quantum kernel classifies those signals better than a conventional support vector machine. The methylation cohort covered 813 individuals with 188 cancers across 56 methylation targets; the fragmentomics cohort covered 718 individuals with 172 cancers across 473 genomic regions. Features were encoded with angle and dense-angle feature maps across circuits of 10 to 20 qubits, using three entanglement patterns, and the resulting fidelity kernels fed a precomputed-kernel support vector machine and a kernel-PCA logistic regression.
A split verdict, measured in simulation
On fragmentomics the quantum-kernel models reached held-out AUC values around 81 to 82 percent against a classical baseline near 78 to 79 percent. On methylation the ordering reversed, with the classical model centered around 83 to 84 percent and the quantum models between 80 and 82.5 percent. Widening from 20 to 40 features did not reliably help. Every kernel was computed by exact statevector simulation on ordinary computers, so no quantum processor was involved, and the authors list hardware execution under finite sampling and device noise as future work. Quentir places the work at TRL 3 of 9 on the quantum computing pillar and reads the split, rather than either half of it, as the finding.
What Quantum-Inspired Buys a Federated ECG Classifier
A compact model for hospitals that cannot pool their ECGs
Hospitals, clinics and wearable makers each hold electrocardiograms they cannot simply pool, so collaborative model training moves the model instead of the data. A new preprint evaluates a hybrid quantum-inspired Kolmogorov-Arnold network for arrhythmia classification under federated averaging, where every site trains locally and transmits only model updates. Because each round's cost scales with model size, a smaller network makes the whole federation cheaper to run for every participant.
Measured savings, with clearly stated edges
On the public MIT-BIH benchmark the network used 37.35 percent fewer trainable parameters and cut communication cost by 24.89 percent; on the INCART dataset the reductions reached 44.81 and 36.41 percent, while most aggregate and minority-class metrics matched or improved on the baseline. The comparison runs against a plain multilayer perceptron on retrospective public datasets, and the federation is simulated, so the result marks a design direction rather than a clinical capability.
What the quantum label does and does not mean
The architecture borrows its learnable functions from quantum machine learning, in the form of single-qubit data re-uploading circuits, yet every calculation runs on classical computers. Quentir places this work at TRL 3 of 9 with the quantum pillar explicitly not applicable: functioning software on recorded data, no quantum processor anywhere in the loop. The honest summary is that federated ECG learning gained a smaller, cheaper collaborative classifier from quantum-derived mathematics, and the evidence record should carry it under exactly that description.
Forty-Six Qubits, One Small Cancer Dataset
What the 46-qubit result did
A cancer neoantigen may differ from an ordinary human peptide by one amino acid. The peptide must bind to a patient’s HLA molecule and then be recognized by a T cell. A Science Advances team used quantum convolutional neural networks to model those two filters and combined them in Q-CHIPP.
The largest hardware experiment represented a full nine-amino-acid peptide with 46 qubits. It used 150 training peptides and 50 test peptides, with 20,000 shots per circuit and two noise-mitigation methods. The model reached F1 0.70. The paper’s unrestricted classical network scored 0.65 and its random forest 0.66 on larger training and test splits.
The biological task is harder than binding
A peptide that binds to HLA may still provoke no T-cell response. Q-CHIPP therefore combines a binding model with an immunogenicity model trained only on confirmed binders. This design addresses a known confound: mixing binders and nonbinders can make binding performance look like immunogenicity prediction.
The patient result remains retrospective
The researchers applied Q-CHIPP to 209,889 candidate peptides from 111 people with HLA-A*02:01-positive lung cancer treated with immunotherapy. Predicted antigen burden was associated with overall survival, but the candidate peptides were not experimentally screened for immunogenicity. The study supports external biological testing, not prospective prediction for a new patient.
Does a Quantum Layer Change What a Medical AI Sees?
One layer makes the comparison unusually clean
A July 2026 preprint compares two medical-image classifiers that share the same convolutional backbone and a comparable number of trainable parameters. They differ in one intermediate layer. One branch uses a dense classical layer; the other uses a four-qubit circuit emulated on a conventional computer. This parameter-matched comparison helps isolate the contribution of the circuit-shaped representation without changing the rest of the network.
The relative gain appeared between small and large datasets
The models classified retinal optical coherence tomography images, with training sets ranging from 200 to 30,000 examples. The hybrid model performed best relative to the classical comparator around 800 and 2,000 training images, an intermediate-data regime. At the largest sizes, the classical CNN moved ahead and reached the study's highest retinal test accuracy, 93.7 percent. A robustness check using two-dimensional slices from an OASIS-1 dementia MRI dataset followed the same broad pattern: the hybrid advantage narrowed with more data, then reversed slightly.
The attention maps changed with sample size
The researchers also compared SHAP attention maps, which estimate where image pixels influence a model's output. With 1,000 retinal images, the hybrid maps appeared more concentrated around retinal layers and disease-related structures, while the classical maps were more fragmented in the examples shown. At 30,000 images, the models' highlighted regions overlapped much more. These maps do not establish clinically validated biomarkers. The study includes no physical quantum-hardware run, prospective clinical test, external hospital validation, clinician reader study, or patient outcome. Its contribution is narrower and useful: a controlled account of when a small classically simulated circuit changed model performance and apparent visual focus, and when additional data favored the matched classical layer.
The Quantum Molecule Generator Hit a Chemistry Limit
A quantum generator enters the chemistry contest
A medicinal chemist can reject a proposed molecule in seconds. A model has a harder job: it must learn which structures are chemically legible, avoid repeating itself, and move toward useful properties without mistaking a score for a drug. A 2023 experiment placed a quantum molecule generator inside that contest. Researchers swapped variational quantum circuits into three parts of a molecular generative adversarial network and compared them on the QM9 dataset.
Fewer discriminator parameters, mixed chemical results
A quantum discriminator used only 50 learnable parameters, compared with about 22,000 in a deliberately reduced classical discriminator. The hybrid system containing the quantum discriminator produced 46.59 percent unique outputs among valid molecules, versus 2.08 percent for that reduced comparator, and had a better distributional-fit score. Its validity was only 31.34 percent, however, compared with 99.78 percent. Larger classical discriminators also surpassed it on uniqueness and distributional fit. In a separate goal-directed test, a quantum noise source improved mean drug-likeness and synthetic-accessibility scores, then produced far fewer valid, nonrepeated outputs. The selected objective rose as the useful range of generated chemistry narrowed.
What the benchmark can and cannot establish
The study is computational. It reports no run of the molecular generator on physical quantum hardware, no synthesized compound, no biological assay, and no drug candidate. Its contribution is architectural: different quantum components produce different bargains among parameter efficiency, property optimization, validity, uniqueness, and similarity to the training distribution. The quantum-generator variant also required roughly 3.5 days per epoch on the reported classical compute instance and struggled to generate unique and valid molecules after ten epochs.
For quantum medicine, that tradeoff matters more than the label on the model. A generator can look efficient while shifting work into chemical filtering. It can score well while returning to the same narrow family of structures. The study's lasting question is therefore precise: which chemical possibilities disappear when a quantum-assisted score improves? Better hardware will change runtime, but it will not answer that measurement question for medicinal chemists.