Quantum-Inspired Orthonormal CNN for Energy-Efficient Medical Image Denoising

Sayantan Dutta
Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, PMLR 315:2097-2117, 2026.

Abstract

Medical imaging modalities (MRI, CT, PET, US) are often degraded by acquisition noise, which obscures subtle anatomical details and compromises diagnostic reliability. Conventional denoising approaches, including spatial filters and deep learning (DL) models, often struggle to balance noise suppression with preservation of fine structures, and state-of-the-art architectures typically incur high computational and energy costs. This work introduces a novel quantum-inspired convolutional neural network (QICNN) that embeds principles of orthonormal basis representation and unitary channel mixing into a compact UNet-style architecture. By constraining convolutional kernels to orthonormal subspaces and enforcing norm-preserving transformations, QICNN eliminates feature redundancy, stabilizes optimization, and maintains energy consistency across layers. Evaluations on real noisy brain MRI datasets show that QICNN achieves superior texture fidelity and lesion conspicuity compared to standard DL models, as evidenced by improvements in GLCM-based metrics and contrast-to-noise ratio. In addition to quality gains, QICNN reduces parameter count by $\sim$93%, inference latency by $\sim$98%, and energy consumption by $\sim$97% relative to transformer-scale denoisers, significantly lowering computational overhead and carbon footprint. These findings highlight the potential of physics-guided design to deliver interpretable, efficient, and clinically robust solutions for medical image restoration.

Cite this Paper

BibTeX
@InProceedings{pmlr-v315-dutta26a, title = {Quantum-Inspired Orthonormal CNN for Energy-Efficient Medical Image Denoising}, author = {Dutta, Sayantan}, booktitle = {Proceedings of The 9th International Conference on Medical Imaging with Deep Learning}, pages = {2097--2117}, year = {2026}, editor = {Huo, Yuankai and Gao, Mingchen and Kuo, Chang-Fu and Jin, Yueming and Deng, Ruining}, volume = {315}, series = {Proceedings of Machine Learning Research}, month = {08--10 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v315/main/assets/dutta26a/dutta26a.pdf}, url = {https://proceedings.mlr.press/v315/dutta26a.html}, abstract = {Medical imaging modalities (MRI, CT, PET, US) are often degraded by acquisition noise, which obscures subtle anatomical details and compromises diagnostic reliability. Conventional denoising approaches, including spatial filters and deep learning (DL) models, often struggle to balance noise suppression with preservation of fine structures, and state-of-the-art architectures typically incur high computational and energy costs. This work introduces a novel quantum-inspired convolutional neural network (QICNN) that embeds principles of orthonormal basis representation and unitary channel mixing into a compact UNet-style architecture. By constraining convolutional kernels to orthonormal subspaces and enforcing norm-preserving transformations, QICNN eliminates feature redundancy, stabilizes optimization, and maintains energy consistency across layers. Evaluations on real noisy brain MRI datasets show that QICNN achieves superior texture fidelity and lesion conspicuity compared to standard DL models, as evidenced by improvements in GLCM-based metrics and contrast-to-noise ratio. In addition to quality gains, QICNN reduces parameter count by $\sim$93%, inference latency by $\sim$98%, and energy consumption by $\sim$97% relative to transformer-scale denoisers, significantly lowering computational overhead and carbon footprint. These findings highlight the potential of physics-guided design to deliver interpretable, efficient, and clinically robust solutions for medical image restoration.} }
Endnote
%0 Conference Paper %T Quantum-Inspired Orthonormal CNN for Energy-Efficient Medical Image Denoising %A Sayantan Dutta %B Proceedings of The 9th International Conference on Medical Imaging with Deep Learning %C Proceedings of Machine Learning Research %D 2026 %E Yuankai Huo %E Mingchen Gao %E Chang-Fu Kuo %E Yueming Jin %E Ruining Deng %F pmlr-v315-dutta26a %I PMLR %P 2097--2117 %U https://proceedings.mlr.press/v315/dutta26a.html %V 315 %X Medical imaging modalities (MRI, CT, PET, US) are often degraded by acquisition noise, which obscures subtle anatomical details and compromises diagnostic reliability. Conventional denoising approaches, including spatial filters and deep learning (DL) models, often struggle to balance noise suppression with preservation of fine structures, and state-of-the-art architectures typically incur high computational and energy costs. This work introduces a novel quantum-inspired convolutional neural network (QICNN) that embeds principles of orthonormal basis representation and unitary channel mixing into a compact UNet-style architecture. By constraining convolutional kernels to orthonormal subspaces and enforcing norm-preserving transformations, QICNN eliminates feature redundancy, stabilizes optimization, and maintains energy consistency across layers. Evaluations on real noisy brain MRI datasets show that QICNN achieves superior texture fidelity and lesion conspicuity compared to standard DL models, as evidenced by improvements in GLCM-based metrics and contrast-to-noise ratio. In addition to quality gains, QICNN reduces parameter count by $\sim$93%, inference latency by $\sim$98%, and energy consumption by $\sim$97% relative to transformer-scale denoisers, significantly lowering computational overhead and carbon footprint. These findings highlight the potential of physics-guided design to deliver interpretable, efficient, and clinically robust solutions for medical image restoration.
APA
Dutta, S.. (2026). Quantum-Inspired Orthonormal CNN for Energy-Efficient Medical Image Denoising. Proceedings of The 9th International Conference on Medical Imaging with Deep Learning, in Proceedings of Machine Learning Research 315:2097-2117 Available from https://proceedings.mlr.press/v315/dutta26a.html.

Related Material