In an article recently published in the journal Bioengineering, researchers proposed a novel hybrid quantum framework for detecting lung cancer.
The QCNN architecture with quantum operations and measurements. Image Credit: https://www.mdpi.com/2306-5354/11/8/799
Background
Lung cancer, one of the most common types of cancer around the world, presents substantial health challenges. Thus, early detection of this disease is essential to simplify treatment and improve patient outcomes. Both chest X-rays (CXR) and computed tomography (CT) scans are crucial tools for lung cancer detection.
Although CXR radiographs or CT scans are currently utilized for lung tumor diagnosis, neither provides a holistic understanding of the diversity and complexity of these tumors. The manual identification of tumors is error-prone, inconsistent, and challenging. Specifically, the radiologist's response in identifying tumors will vary based on the radiologist's expertise and the imaging technique's prominence.
Tumors can be identified more precisely, quickly, and objectively from different images using automated methods, specifically deep learning (DL) models. In medical imaging like CXR and CT, DL methods have recently been employed for lung tumor classification.
However, DL networks often require extended computation times and significant computing power to process data. Additionally, their performance primarily depends on the data size and network hyperparameter precision. Thus, misconfigured hyperparameters can significantly diminish a DL model's efficiency, robustness, reliability, and accuracy.
The recent advancements in quantum computing can address these challenges, improving the scalability, accuracy, and speed of DL models. These methods enhance diagnostic accuracy and robustness and accelerate the processing speeds of DL systems through efficient allocation of computation resources.
The Study
In this work, researchers introduced a novel hybrid architecture that combines DL with quantum computing to enhance lung cancer detection accuracy using both CT images and CXR. The proposed framework leverages quantum circuits for classification and pre-trained models for feature extraction, specifically targeting the improved classification of lung tumors.
The study aimed to overcome the limitations of existing methods in distinguishing benign from malignant lung tumors. To achieve this, the researchers employed pre-trained transfer learning (TL) models, fine-tuned for CT and CXR image-based lung tumor classification. A hybrid quantum layer was also incorporated, which combined features from both CXR and CT images, further enhancing the TL model's classification performance.
The framework was evaluated using two widely recognized open-source datasets: the Lung Image Database Consortium image collection (LIDC-IDRI) and ChestX-ray8. These datasets are extensively used in lung disease research, providing a robust basis for evaluating the proposed system.
Methodology
The proposed system consisted of three key modules: image acquisition, TL model tuning, and quantum classification and learning. These modules worked in tandem to improve lung disease classification.
-
Image Acquisition and Pre-processing: The process began with the acquisition and pre-processing of extensive medical image datasets to ensure uniformity and quality. This step was crucial for effective model training and evaluation.
-
TL Model Tuning: Pre-trained models, such as RepVGG, ResNet50-V2, and VGG16, were fine-tuned to specialize in lung disease classification tasks. These models served as the backbone for feature extraction in the hybrid system.
-
Quantum Classification and Learning: Quantum Convolutional Neural Networks (QCNNs) were developed and integrated with the fine-tuned TL models. This integration created a hybrid system that leveraged the strengths of both classical and quantum computing. The hybrid models were then optimized, trained, and evaluated based on performance metrics such as specificity, sensitivity, and accuracy.
The final optimized model was prepared for deployment in clinical settings, ensuring compatibility with existing medical systems and scalability for broader use. This approach not only improves classification accuracy but also paves the way for more effective and scalable lung cancer detection solutions using advanced computational techniques.
Significance of this Study
The proposed hybrid quantum framework achieved state-of-the-art performance in classifying lung tumors, outperforming methods that rely solely on CT or CXR images and conventional machine-learning techniques.
Notably, the RepVGG model, when integrated with the hybrid quantum layer, achieved a classification accuracy of over 92 %, surpassing the accuracy of other standard methods by more than 3 %. In addition to its impressive accuracy, the system excelled in other critical performance metrics, attaining 94 % sensitivity, 92 % precision, a 93 % F1-score, and 90 % specificity.
The integration of quantum computing not only enhanced the model's scalability and processing speed but also positioned the system as a highly effective tool for early lung cancer diagnosis and screening. The system required just 2.32 hours of computational training time to achieve superior accuracy, compared to methods like Quanvolution, which required 2.45 hours to reach 88.24 % accuracy.
In summary, the findings underscore that this hybrid approach can identify lung cancer signatures with greater accuracy, speed, and efficiency than conventional methods. This highlights the potential of hybrid computational technologies to revolutionize early cancer detection and pave the way for broader clinical applications and improved patient care outcomes.
However, it is important to note that certain types of lung cancer, due to their unique molecular or morphological characteristics, may not be suitable for this framework, indicating the need for further refinement and customization in those cases.
Journal Reference
Martis, J. E., M S, S., R, B., Mutawa, A. M., Murugappan, M. (2024). Novel Hybrid Quantum Architecture-Based Lung Cancer Detection Using Chest Radiograph and Computerized Tomography Images. Bioengineering, 11(8), 799. DOI: 10.3390/bioengineering11080799, https://www.mdpi.com/2306-5354/11/8/799
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.