Quantum-classical hybrid pipeline boosts polyp detection with cold-atom reservoir computing

A team of researchers led by Nuno Batista has proposed a novel hybrid quantum-classical approach for medical image classification, focusing on binary polyp detection. Their pipeline leverages neutral-atom reservoir computing, a quantum machine learning technique that uses the natural dynamics of cold atoms as a computational resource. To reduce the dimensionality of medical images, they employ a guided auto-encoder that learns compact, discriminative latent representations. These representations are then encoded as pulse detuning parameters in a Rydberg Hamiltonian, and quantum embeddings are obtained through expectation values before being passed to a linear classifier.

The key innovation is overcoming the 'gradient barrier' caused by non-differentiable quantum measurements. The team introduced a differentiable surrogate model that emulates the quantum layer, enabling standard backpropagation through the entire pipeline. The training is jointly optimized for classification accuracy and faithful image reconstruction from the auto-encoder. Simulations show that this method significantly outperforms traditional approaches using PCA or unguided auto-encoders. The paper, accepted at the 2025 IEEE International Conference on Quantum AI, demonstrates robustness through ablation studies and highlights the feasibility of quantum-enhanced medical imaging even in the current NISQ era.