Integrating finite element analysis and machine learning for non-invasive tumor detection: a piezoelectric tactile sensor-based vibration absorber approach

Abstract Timely cancer detection is crucial for reducing mortality rates associated with delayed diagnosis and treatment. This study presents a novel, non-invasive tumor detection technique that integrates finite element analysis (FEA) with machine learning (ML) models. A piezoelectric sensor, developed using vibration absorber phenomena, was designed to detect changes in tissue stiffness caused by tumors. The sensor's interaction with soft tissue of varying Young’s modulus (9 to 185 kPa) was simulated to mimic the properties of cancerous tissue. Modal and harmonic analyses using ANSYS software, including indentation trials, were conducted to study the sensor’s response to stiffness variations, creating a dataset based on tumor sizes (5, 10, 12, 15, 17, 20, and 25 mm in diameter) at different vertical and horizontal distances from the tissue surface to the tumor center, simulating various growth stages and depths. The collected data were used to train two machine learning models: a recurrent neural network (RNN) for precise tumor localization and sizing and a feedforward neural network (FNN) to estimate tissue stiffness through sensor absorber frequencies. The proposed methodology demonstrated promising results, with a minimum error of 0.04 mm in tumor size estimation and 0.0319 kPa in stiffness detection. This approach offers potential improvements in early tumor detection by providing accurate and noninvasive diagnostics, particularly in resource-constrained environments.