International Journal of Biomedical Engineering and Technology (3 papers in press)

Regular Issues

• TriFormer-Seg: Leveraging Multi-Scale Vision Transformers and CNNs for Clinically Accurate Breast Cancer Segmentation  
  by Sri Vidhya Komarina, Ravi Kumar Jatoth 
  Abstract: This study introduces TriFormer-Seg, a hybrid deep learning model designed to perform accurate semantic segmentation of breast cancer in both MRI and histopathology images. The architecture leverages the distinct advantages of three complementary components: vision transformers for global contextual understanding, Swin transformers for multi-scale hierarchical representation, and convolutional neural networks (CNNs) for extracting fine-grained local details. By combining these methods, the model addresses the complex nature of medical imaging, where capturing both broad structural patterns and precise boundary details is essential. A streamlined decoding pathway, enhanced with boundary refinement, improves edge accuracy, while a specialised multi-scale fusion module harmonises information across different feature levels. Evaluated on datasets comprising dynamic contrast-enhanced MRI and histopathological images, TriFormer-Seg demonstrates consistent improvements over leading segmentation techniques, achieving Dice coefficients of up to 89.25% on histopathology images and 0.82 on breast DCE-MRI. Visual comparisons further confirm the models ability to generate segmentation masks closely aligned with expert annotations, indicating strong performance within the evaluated datasets under varying tissue appearance and imaging conditions.
  Keywords: Breast cancer; Semantic segmentation; Histopathology; Transformers; Deep learning; MRI Image.
  DOI: 10.1504/IJBET.2026.10080415
   
  
• DHA-max Net: Deep high-order Attention maxout Network for Age-related Macular Degeneration Detection using OCT images  
  by Rahul Sukumar Nejkar, Shabnam Farook Sayyad, Gotekar Rajesh Balkrishn 
  Abstract: Nowadays, Age-related Macular Degeneration (AMD) is the major cause of blindness and severe visual impairment in elderly people worldwide. In this paper, a Deep High-order Attention maxout Network_Fractional Political Flower Pollination Optimization (DHA-max Net_FPFPO) is designed for detecting AMD from Optical Coherence Tomography (OCT) images. Initially, the input image is acquired, and the segmentation is performed using Ensemble U-Net with the hyperparameters optimally adjusted using Political Flower Pollination Optimization. Later, feature extraction is performed, and the AMD is finally detected by DHA-max Net. Here, the detection performance is improved by optimally adjusting the hyperparameters of DHA-max Net using the Fractional Political Flower Pollination Optimization (FPFPO). The experimental outcomes of DHA-max Net_FPFPO are validated, where DHA-max Net_FPFPO attained high performance with a high specificity of 94.8%, sensitivity of 98.5%, accuracy of 96.6% and precision of 95%.
  Keywords: Optimization; Deep learning; Optical Coherence Tomography; Ensemble U-Net; Age-related Macular Degeneration.
  DOI: 10.1504/IJBET.2026.10081250
   
  
• Ensemble Learning with Fuzzy Attention SegNet and nnU-Net Using Log- Cosh Softmax Loss for Optimal Brain Tumor Segmentation in MRI  
  by Jyothi C. K, Anupama S. Awati, Dattaprasad A. Torse 
  Abstract: Brain tumor is an unregulated proliferation of aberrant cells that impair the normal function of the human brain. The segmentation of brain tumor plays a vital role in brain tumor detection. Various techniques are currently used for brain tumor segmentation; however, these methods face challenges in capturing comprehensive long-range dependencies within the feature space. In this research, the optimization-based ensemble deep learning approach Fuzzy Attention SegNet+no new-UNet_Serval Ablation Optimization(FA-SegNet+nnU-Net_SeAO) is developed to segment brain tumor regions from Magnetic Resonance Imaging (MRI) scans. The Attention-Guided Convolutional Neural Network (ADNet) is utilized to denoise the input image. Then, the denoised tumor region in the brain is segmented through an ensemble of Fuzzy Attention SegNet (FA-SegNet) and no new-Net (nnU-Net), trained independently using Serval Ablation Optimization (SeAO). Furthermore, the efficacy of FA-SegNet+nnU-Net_SeAO is investigated, where the model attained segmentation accuracy of 94.766%, Intersection over Union (IoU) of 95.766%, and dice coefficient of 96.766%.
  Keywords: Brain tumor segmentation; Deep learning; Magnetic Resonance Imaging; Optimization; Fuzzy Attention SegNet.
  DOI: 10.1504/IJBET.2026.10081721