Tissue Individual Signatures through Machine Learning Analysis of Ultrasound Images

Advancements in sensors, computing, and surgical data science have significantly enhanced minimally invasive procedure, offering benefits such as reduced trauma, faster recovery, and lower infection risk. Despite these improvements, challenges persist regarding intraoperative accuracy, safety, and localization of surgical tools. Ultrasound imaging, while widely used for real-time visualization, faces limitations such as artifacts and poor alignment with instruments during procedures. This study presents a preliminary framework for ultrasound-based classification of materials with tissue-like properties using machine learning algorithms. Artificial phantoms constructed from accessible materials and biological tissues were imaged with ultrasound. Extracted image features were preprocessed using Gabor, Tamura, and Hessian filters, followed by statistical parameterization. Classification models including Random Forest, Support Vector Machines, and Discriminant Analysis were evaluated using the F1 score, achieving up to 0.8 in classification performance. The results indicate that ultrasound imaging, combined with statistical learning, has the potential to differentiate materials with tissue-relevant properties. This approach may contribute to enhanced intraoperative navigation and real-time tissue characterization. Future work will include dataset expansion, statistical validation, and integration with clinical systems.