International Journal of Computer Sciences and Engineering

ROI-Aware Quantum Autoencoder Framework for Efficient Medical Image Compression

Gangadhar Durgada, Surekha Samsani

Published Aug 31, 2026• 0 citations

Abstract

Medical imaging systems generate large volumes of high-resolution images, making efficient storage, transmission, and real-time clinical analysis challenging. This work proposes a Region of Interest (ROI)-Aware Quantum Autoencoder (ROI-QAE) framework to achieve high compression efficiency while preserving diagnostic image quality. The ROI-Aware Quantum Autoencoder (ROI-QAE) is a four-stage pipeline for compressing computed tomography (CT), magnetic resonance imaging (MRI), and COVID-19 chest X-ray images within a single unified implementation. The first stage uses a Vision Transformer (ViT) with eight-head multi-head self-attention over 32×32-pixel patches to generate a binary region-of-interest (ROI) mask, requiring no labelled annotations. The second stage applies an ROI-aware adaptive JPEG engine. The third stage introduces a Quantum Autoencoder (QAE) built in PennyLane that operates on 8×8-pixel patches. A three-layer classical encoder maps each 64-dimensional patch to four qubit angle values, a three-layer Strongly Entangling Layers parameterized quantum circuit (PQC) on a lightning Qubit device processes these angles, and a three-layer classical decoder reconstructs 64-pixel values. The fourth stage applies lossless Huffman entropy coding. Three modality-specific QAE models are trained on 60,000 patches each (300 training images × 200 patches per image) using the Adam optimizer (initial learning rate 5×10⁻³, weight decay 1×10⁻⁵) with a cosine-annealed learning-rate schedule and gradient-norm clipping. Evaluated on 50 held-out test images per modality (drawn from pools of 120 CT, 120 MRI, and 100 X-ray test images), the framework yields mean PSNR of 23.70 dB, 24.73 dB, and 29.18 dB; SSIM of 0.7969, 0.7256, and 0.7396; and space savings of 81.6%, 81.2%, and 80.5% for CT, MRI, and X-ray respectively. Huffman compression ratios reach 1.7304 (CT), 1.3544 (MRI), and 1.1470 (X-Ray). The framework demonstrates that parameterized quantum circuits with as few as 36 quantum parameters can be embedded into a clinically aware compression pipeline and trained on commodity cloud hardware.

DOI: 10.26438/ijcse.v14i8.7509

Related papers

A Quantum-Enhanced Artificial Neural Network Model for Efficient Medical Image Compression

Balasubramani Subbiyan, P. N. Renjith, Kavisankar Leelasankar +3 more

The ability to effectively store and transmit high-resolution images such as MRI and CT scans without losing quality is critical to modernizing medical imaging. Traditional compression methods risk losing essential medical image data, which requires perfect detail for diagnosis. Quantum algorithms use superposition and entanglement to compress faster while preserving important information. This research presents a Quantum-enhanced Artificial Neural Network (QANN) model that combines quantum feature extraction with classical neural network topologies to improve image compression. Our approach consists of converting standardized classical data into quantum states, controlling these states using parameterized quantum circuits, and measuring the resulting states to produce enhanced feature vectors. The quantum-enhanced features are fed into a traditional neural network for image compression. The experimental results clearly show that our QANN framework outperforms standard models in terms of accurate reconstructed images, reduced size, and increased space-saving percentage, especially when dealing with large and complicated datasets. The QANN model demonstrates how quantum computing can significantly enhance the effectiveness of medical image processing solutions. Kaggle brain CT and MRI datasets and COVID-CXNet chest x-ray images are used. The proposed QANN model improves peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Using quantum technology, the image size is reduced for MRI (73.3 %), X-ray (74.1%), and CT-SCAN (71.8%) to save space.

IEEE AccessJan 1, 202525 citations

Compression-Driven Anomaly Detection in Brain MRI Using an Interpretable Quantum Autoencoder

Santanu Ganguly, Xing Liang, Dimitrios Makris

We study a quantum autoencoder (QAE) for compression-driven anomaly detection in brain MRI data. The approach leverages angle encoding to map image patches into quantum states, followed by a variational encoder-decoder architecture trained to discard information via auxiliary trash qubits. Anomaly scores reflect the degree to which inputs resist compression relative to normal data, with higher scores corresponding to deviations from the learned normal manifold. Evaluated on publicly available brain MRI DICOM datasets, the method achieves a slice-level ROC-AUC of approximately 0.95 and a patch-level ROC-AUC of approximately 0.813, outperforming classical autoencoder and PCA baselines. Analysis of the learned parameters reveals a pronounced encoder-decoder asymmetry, where effective anomaly detection arises from structured information compression within the encoder rather than increased parameter magnitude or decoder expressivity. This results in a controlled compression-reconstruction trade-off with a clear operating regime that supports principled threshold selection. Qualitative evaluation further shows that the QAE produces spatially localized anomaly heatmaps aligned with tumorous regions. The results, supported by promising baseline performances, demonstrate that quantum autoencoders provide an interpretable and controllable mechanism for anomaly detection based on incompressibility with respect to a learned latent representation. This work highlights the potential of quantum autoencoders as a principled tool for studying compression dynamics in quantum machine learning, with promising implications for decision support in medical imaging workflows.

quant-pharXiv:2606.27411Jun 26, 20260 citations

High-Fidelity ROI CT Reconstruction with Limited Quantum Resources via Hybrid Classical-Quantum Refinement

Hyunju Lee, Jeonghwa Lee, Kyungtaek Jun

Quantum optimization for computed tomography (CT) reconstruction is constrained by the number of binary variables required for image representation, making direct whole-image quantum reconstruction difficult for large or structurally complex objects. We propose a hybrid region-of-interest (ROI) refinement framework in which a coarse global image is first reconstructed by quantum tomographic reconstruction (QTR) and quantum compressed sensing tomographic reconstruction (QCSTR), filtered backprojection (FBP), or simultaneous algebraic reconstruction technique (SART), and quantum optimization is then applied only to the selected ROI through a residual projection-image formulation. This strategy reduces the effective QUBO size while preserving high-fidelity reconstruction in the target region. Experiments on three discrete phantom samples show that both QTR/QCSTR+QTR/QCSTR and SART+QTR/QCSTR achieve accurate ROI reconstruction for moderate-size cases under a reduced-angle setting. For the largest and most complex sample, the quality of the coarse global estimate becomes critical, and the best result is obtained when a stable classical coarse reconstruction is combined with second-stage ROI-only QTR/QCSTR. Among the tested pipelines, SART+QTR/QCSTR achieves the lowest average ROI RMSE and MAE. The results indicate that the practical advantage of quantum-assisted CT reconstruction lies in reserving quantum optimization for local refinement while using classical reconstruction to stabilize the global background.

quant-pharXiv:2605.29472May 29, 20260 citations

Quantum-Enhanced AI Framework for Classification of Encryption Techniques in Medical Image ROIs for Secure Transmission

Anindita Das Badhan, Diganth Vijayan R, K. Ashwini

Secure transmission of encrypted medical images demands trustworthy identification of the used encryption method, especially in settings where the availability of metadata is uncertain or unreliable. In this paper, a quantum-enhanced AI framework for the classification of encryption methods used in medical image Regions of Interest (ROIs) is introduced. After the extraction of the ROI and chaotic map-based encryption, a hybrid quantum-classical learning strategy is used to classify the encryption method without depending on any metadata. The proposed quantum AI framework combines variational quantum circuits as efficient latent feature encoders in CNN architectures. The high-dimensional CNN features are then compressed into low-dimensional latent spaces before the quantum embedding process, enabling the classification task to be conducted on the highly compressed feature representation. Quantum-enhanced ResNet18, VGG11, and EfficientNet-B0 models were trained and tested on encrypted ROIs extracted from dermoscopic medical images. The experimental results show that the proposed framework can successfully distinguish between different encryption methods, confirming the feasibility of quantum-assisted cipher classification for secure transmission of medical images. This work shows the potential of quantum-enhanced feature representation in privacy-preserving healthcare applications and automatic decryption routing in medical systems.

International Conference on Networking and Communications (ICNWC)Apr 6, 20260 citations

Quantum autoencoder for image compression through reducing qubits usage

Yuxing Wei, Hai-Sheng Li, Cong Hu

Quantum autoencoder, as a quantum machine learning algorithm, provides an important research direction for image compression. This paper proposes a quantum autoencoder specifically designed for image compression. We first use feature mapping to store images as quantum states and propose a new parameterized quantum circuit and a pre-training method. After pre-training, a reference state is determined and the circuit parameters are updated through further training. Finally, we experimentally validate the quantum autoencoder. For reconstructed 10-qubit images, the fidelity reaches 0.98 and the SSIM reaches 0.927. For reconstructed 10-qubit noisy images, the PSNR improves by 1.4[Formula: see text]dB. Simulation results demonstrate that the proposed quantum autoencoder efficiently compresses quantum images while also possessing denoising capabilities.

International Journal of Quantum InformationOct 30, 20250 citations

Adaptive Hybrid Quantum Image Representation for Efficient Encoding of Medical and SAR Image

Thouseef f

Abstract Quantum image representation constitutes a foundational component of quantum image processing; however, its practical deployment on Noisy Intermediate-Scale Quantum (NISQ) hardware is fundamentally constrained by the cost of quantum state preparation. Existing representations such as the Flexible Representation of Quantum Images (FRQI) and the Novel Enhanced Quantum Representation (NEQR) rely on uniform, full-resolution encoding strategies that incur rapidly increasing gate counts and circuit depths, leading to severe decoherence and error accumulation on near-term devices. These limitations are particularly pronounced for heterogeneous and information-sparse modalities such as medical Magnetic Resonance Imaging (MRI) and Synthetic Aperture Radar (SAR), where uniform pixel-wise encoding introduces substantial redundant state preparation. To address these constraints, we propose Adaptive Hybrid Quantum Image Representation (AHQIR), a saliency-driven selective quantum state-preparation framework designed for efficient encoding of medical and SAR imagery under NISQ limitations. AHQIR departs from global encoding by employing a classical adaptive preprocessing stage to identify diagnostically significant regions in MRI and dominant backscatter structures in SAR images using a statistical thresholding criterion, thereby restricting quantum encoding to a salient pixel subset S. The quantum realization of AHQIR integrates Perception-Aided Encoding (PE) and Coherent-Size Encoding (CE) mechanisms, which condition state preparation on salient pixel coordinates and eliminate the requirement for power-of-two image padding. As a result, circuit depth and multi-qubit gate count scale with the number of salient pixels rather than total image size, enabling physically realizable circuits within current coherence limits. This selective and intentionally lossy design sacrifices global pixel-level fidelity in non-salient regions in order to preserve semantic and structural fidelity within regions of interest. AHQIR is evaluated against multiple state-of-the-art quantum image encoding schemes, including FRQI and NEQR, across Brain Tumor MRI and real-world SAR datasets. Experimental results demonstrate substantial reductions in gate count and improved robustness under realistic noise models, while maintaining reconstruction errors below clinically and analytically relevant thresholds for salient regions. These findings indicate that saliency-aware, selective state preparation offers a practical and scalable pathway for quantum.

Mar 31, 20260 citations