Enhancing Intra-Body Localization for Medical Procedures: A Hybrid Approach Integrating Mathematical Models and Machine Learning

M Souganttika

This dissertation tackles important gaps in intra-body position estimation, which are vital for improving medical interventions. The project intends to create robust, precise, and efficient real-time position estimating systems by combining mathematical models and machine learning approaches, specifically in healthcare. Position estimation is critical in clinical operations that have historically been performed manually, which are time-consuming and prone to mistakes due to their dependence on trial-and-error techniques. Existing techniques, whether based on mathematical models or machine learning, often work independently, focusing on exploiting current data or enhancing accuracy through modeling. This research proposes unique hybrid techniques that combine mathematical models and machine learning to create ‘experienced’ algorithms. These algorithms are intended to emulate the decision-making abilities of experienced doctors, improving accuracy and efficiency in real-time medical localization systems. The study addresses the broad spectrum and complexity of medical localization systems and their requirements by focusing on specific medical procedures that require precision, such as placing electroencephalography (EEG) electrodes, inserting nasogastric Tubes (NGTs), and identifying Sentinel Lymph Nodes (SLN) during biopsies. Distinct mathematical and hybrid models for each procedure developed during research ensure the enhancement in accuracy in the position estimates, accommodating to the dynamics of human anatomy while also fulfilling the system requirements. Integrating these models into the position estimation system instills a sense of ‘experience’ in the system, reflecting years of acquired skill. Extensive testing across various circumstances has revealed the potential of hybrid techniques for enhancing the accuracy of real-time position estimating systems in healthcare. These findings highlight a significant improvement over traditional methods relying solely on empirical data or purely mathematical models. Specifically, the hybrid approach developed for predicting optimal EEG electrode position estimations is approximately 20% more accurate than those derived from purely mathematical models and about 83% more accurate compared to solely machine learning-based models. Furthermore, the algorithm designed to optimize the insertion of NGTs has shown a 33% reduction in the computational load and cost by reducing the number of sensors used without compromising the positional accuracy (within 5%). This indicates a significant possibility of improving the efficiency and accuracy of medical procedures using the suggested procedure-specific hybrid modeling techniques.

Enhancing Intra-Body Localization for Medical Procedures: A Hybrid Approach Integrating Mathematical Models and Machine Learning

(1)