Quantum Kernelized DeepVision: A Robust 3D U-Net framework for Early Detection of Primary Hyperoxaluria Type 1 in Renal MRIDurga Prasad Kavadi a, Palacharla Ravi Kumar b,*, Dulam Devee Sivaprasad c, Sai Babu VeesamLeave this area blank for abstract info. Quantum Kernelized DeepVision: A Robust 3D U-Net framework for Early Detection of Primary Hyperoxaluria Type 1 in Renal MRIDurga Prasad Kavadi a , Palacharla Ravi Kumar b, *, Dulam Devee Sivaprasad c , Sai Babu VeesamLeave this area blank for abstract info.Propose a novel Quantum Kernelized DeepVision framework integrating quantum kernel embeddings with 3D U-Net to enable robust, anatomically consistent, topologically validated, and interpretable voxel-level detection of Primary Hyperoxaluria Type 1 in renal MRI, achieving significant improvements in localization accuracy, structural consistency, and entropy preservation over classical deep learning models. The early and accurate recognition of Primary Hyperoxaluria Type 1 can be defined as a clinical challenge due to its subtle presentation in renal MRI and the pathological condition being rare with very little labeled data. The available Deep Learning imaging solutions, mostly relying on classical CNNs, generally do not generalize well for high-dimensional changes in structure due to little handling and robustness towards rare patterns. Lack of anatomical consistency, topological validation, and interpretability, which are critical in rare disease diagnosis, are major hindrances for such classical models. To surmount these, we propose a completely novel Quantum Kernelized DeepVision, a mix of the quantum kernel embedding-enhanced 3D U-Net architecture to carry out voxel-level detection of oxalate crystal deposits. The quantum kernel module enriches the representation of the latent space with the Hilbert space succession projection which captures the complex anatomical and spatial dependence. Five new sets of analytical validation and implementation methods are to be developed. (1) Quantum Structural Embedding Consistency assures that learned embeddings are anatomically aligned. (2) Quantum Adversarial Perturbation Probing is an index of robustness of the latent space against structured quantum gradient attacks. (3) Topological Renal Manifold Alignment establishes validation of topological preservation via persistent homology. (4) Quantum-Driven Entropy Projection Analysis quantifies regional information preservation via von Neumann entropy. (5) Hybrid Quantum-Disentangled Visual Attribution provides interpretable voxel-wise class attributions by means of disentangled latent subspaces. Our proposed integrated framework has achieved an improvement of about 41% in localization accuracy, over 60% in topological consistency, and over 70% in the reduced deviation of entropy from the classical baselines. This work lays the foundation for the robust, interpretable, and quantum-enhanced deep imaging models for rare renal disease diagnostics.
Kavadi et al. (Tue,) studied this question.