Medical image segmentation using an optimized three-tier quantum convolutional neural network trained with hybrid optimization approach

• Published in: Multimedia tools and applications Vol. 83; no. 13; pp. 38083 - 38108
• Main Authors: Prasad, S. V. S, Rao, B. Chinna, Rao, M. Koteswara, Kumar, K. Ravi, Vara Prasad, Srisailapu D., Ramesh, Chappa
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2024; Springer Nature B.V
• Subjects: Artificial neural networks; Color; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Feature extraction; Image analysis; Image contrast; Image segmentation; Medical imaging; Multimedia Information Systems; Neural networks; Optimization; Search algorithms; Special Purpose and Application-Based Systems; Texture; Weighting functions; Hybrid Optimization Model; Optimized Mask RCNN; Improved LDP; Medical Image Segmentation; Three-Tier Quantum CNN
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-16980-9

Medical image segmentation is a crucial task in medical image analysis. The proposed method for medical image segmentation involves several steps. First, pre-processing techniques such as Gaussian filtering and contrast stretching are applied to the input image. Next, a region of interest (ROI) is identified from the pre-processed image using an optimized mask RCNN, with the weight function of the RCNN optimized via a new hybrid optimization algorithm- Cuckoo-Spider Optimization, combining Cuckoo Search (CS) and Social Spider Optimization (SSO). After ROI identification, feature extraction is performed, including texture features such as Gray-Level Run Length Matrix (GLRLM), Local rotation invariant Texture Pattern (LrTP), and an Augmented Local Directional Pattern (A-LDP) proposed in this work. Additionally, shape features such as area and perimeter, and color features such as color histogram are extracted. Finally, an optimized three-tier quantum convolutional neural network (O-TT-QCNN) is proposed for segmentation, which can handle complex and heterogeneous medical images. The experimental results demonstrate that the proposed method achieves state-of-the-art performance on several benchmark datasets.