Medicine Henry Quentir Medicine Henry Quentir

EigenOrbit and Trieste Dermatologists Use Quantum-Inspired Butterfly Layers to Certify an Alopecia Image Classifier, arXiv 2611.18775, 12 November 2026

What the EigenOrbit and University of Trieste team posted on arXiv on 12 November 2026

Quentir Medicine Monitor

Evidence-based insights for quantum medicine.

Researchers at the Luxembourg company EigenOrbit and the Dermatology Clinic of Hospital Maggiore in Trieste have adapted an image classifier for hair-loss disorders so that it keeps a mathematical proof of stability, known as certifiable robustness, after it is fine-tuned for a new task. Their tool is a set of quantum-inspired butterfly layers, which borrow the structure of a quantum circuit but run on ordinary computers, and they tested it on trichoscopic images, the magnified scalp photographs dermatologists use to diagnose alopecia. On both clinical tasks the adapted model issued certificates for about twice as many test images as two standard fine-tuning methods, at comparable accuracy.

Stress tests of dermatology classifiers have shown that rotating an image, or photographing the same patch of skin a second time, can change a model's prediction. A test-set accuracy figure reports how often a model was right across a whole collection of images and gives no assurance about any single prediction when the photograph changes slightly. The preprint has not been peer reviewed, and all of its clinical images come from one hospital clinic.

The preprint, "Quantum-inspired orthogonal adapters for certifiably robust medical image classification" on arXiv, appeared on 12 November 2026. Its authors are Olga Tsurkan, Laia Domingo and Asel Sagingalieva of EigenOrbit Sàrl in Luxembourg, and Victoria Orlova, Iris Zalaudek and Raffaele Dante Caposiena Caro of the Dermatology Clinic at Hospital Maggiore of Trieste, University of Trieste. The clinical tasks repeat those of an earlier retrospective study by Caposiena Caro, Orlova, Di Meo and Zalaudek in the Journal der Deutschen Dermatologischen Gesellschaft in 2026. The first task is diagnosis across five classes: alopecia areata, frontal fibrosing alopecia with lichen planopilaris, folliculitis decalvans, androgenetic alopecia and a clinically healthy pattern, drawn from images of 143 patients. The second is the activity of alopecia areata, graded as active, inactive or in regrowth, from images of 57 patients. The test sets hold 717 and 240 original photographs.

Read More