فهرست نشریات علمی دسترسی آزاد ایران

   Background and Objectives: Ultrasound imaging is crucial for breast cancer diagnostics but is prone to subjectivity, causing delays and errors. This study aims to improve diagnostic efficiency and reduce reliance on human interpretation by introducing a novel breast tumor diagnosis technique using a squid game optimization-inspired neural network (SGO-NN).   Methods: The proposed method combines a Siamese neural network (SNN) with SGO techniques to detect tumor variations in ultrasound images. Publicly available breast ultrasound images were preprocessed with a Gaussian filter to enhance quality and remove extraneous information. The Adaptive Watershed Transformation (AWT) approach was used for segmentation, and significant texture characteristics were extracted using the gray-level statistical matrix (GLSM) method. The SGO-NN technique was then trained to identify breast tumors.   Results: The SGO-NN method achieved high performance with sensitivity (94.80%), recall (93.50%), accuracy (91.30%), specificity (93.20%), and precision (92.70%). This performance surpassed existing methods, including Convolutional Neural Networks (CNN), Cubic-Support Vector Machine (C-SVM), and Extreme Gradient Boosting (XG-BOOST).   Conclusion: The SGO-NN method accurately recognizes breast cancers from ultrasound images and outperforms existing methods. This advancement shows significant potential for improving diagnostic processes, reducing human error in medical imaging, and broadening the application of automated diagnostics. Future work should focus on improving computational efficiency, enhancing interpretability through explainable AI approaches, and integrating modern imaging technologies for more accurate and accessible diagnosis.