Early diagnosis of skin diseases greatly improves patient outcomes, but subtle symptoms often demand a dermatologist’s trained eye. While computers can assist by learning from high-quality medical images, the scarcity of such data remains a major barrier to fast, accurate diagnoses.
Qingyue Jiao, a first-year graduate student in the Department of Computer Science and Engineering at the University of Notre Dame, and collaborators from Rensselaer Polytechnic Institute, have developed HybridQ—a new kind of machine-learning model that combines classical and quantum computing techniques. HybridQ is the first to generate clinically relevant skin images in full color, rather than grayscale, marking a significant step toward real-world applications of quantum computing in dermatology.
Previous classical machine-learning models need to train on large volumes of labeled images, which are often unavailable in medical settings. Qingyue’s team employs traditional pre-processing algorithms to break down images of actual skin lesions into their individual features, or latent representations.
Quantum generators then use these representations to create new images, while a “discriminator”—a kind of quality control inspector—rejects or accepts the results based on real-world data. Finally, a post-processing technique eliminates noise, or digital distortions, from the images, readying them for inclusion into usable datasets.
In addition to producing more substantial, higher-quality data, the team’s generated images correct for the lack of breadth and variety in current datasets, both in terms of patient population and disease representation. Less demanding computationally than previous data-augmenting techniques, their model is also more accessible to researchers and clinicians with limited resources.
Finally, the team tested HybridQ on an IBM quantum computer, proving its ability to generate quality images on actual hardware.
“As quantum hardware continues to advance, models such as HybridQ will continue to enhance medical imaging, benefitting a wide range of patients and medical facilities,” said Jiao. The team plans to test the potential of hybrid classical-quantum models for cancer detection as well as neurological disease diagnosis.
Their paper, ‘HybridQ: Hybrid Classical-Quantum Generative Adversarial Network for Skin Disease Image Generation,’ appeared in Cornell University’s arXiv, a free distribution service and an open-access archive for scholarly articles in the fields of physics, mathematics, computer science, quantitative biology, quantitative finance, statistics, electrical engineering and systems science, and economics.
—Karla Cruise, Notre Dame Engineering