Depth determination of simulated biological tissue using X-ray radiography and feature extraction techniques: Evaluation with Bi-LSTM neural network

Accurate determination of tissue depth in medical imaging and disease diagnosis is crucial, especially for complex biological structures. Traditional methods often lack the necessary precision for effective diagnosis and treatment planning. This study investigates the determination of heart-like tissue depths using X-ray outputs and advanced feature extraction techniques. Simulated tissues at depths of 5, 10, 15, and 20 cm were analyzed using the Monte Carlo N-Particle Transport Code (MCNPX), with radiographic images captured at 70 keV. Features such as wavelet transform, Gray Level Co-occurrence Matrix (GLCM), Gray Level Run Length Matrix (GLRM), and fractal analysis were extracted. A Bidirectional Long Short-Term Memory (Bi-LSTM) network was used to predict tissue depth, comparing the performance of optimizers including Adam, RMSprop, and SGD. The results showed that the Stochastic Gradient Descent (SGD) optimizer achieved superior prediction accuracy compared to other optimizers. Statistical performance metrics indicated that SGD outperformed its counterparts, showcasing enhanced precision and reliability in predictive modeling of tissue depths. Mean RMSE: 0.21155, Mean MAE: 0.18522, Mean MBE: 0.03400, Mean MRE: 0.01422, Mean MAPE: 0.01422, Mean SMAPE: 0.01429. The findings demonstrate the high accuracy of the Bi-LSTM model in predicting tissue depths from radiographic images. This study represents a significant advancement in medical diagnostics, providing an innovative solution to longstanding challenges. By integrating advanced imaging techniques with machine learning algorithms and leveraging MCNPX for precise simulation, more accurate and reliable diagnostic tools can be developed, ultimately improving patient outcomes. Future research could explore clinical applications of this approach and further refine the models for greater accuracy. Accurate depth determination is crucial not only in medical applications, such as optimizing radiation doses in radiotherapy, but also in various industrial contexts, such as non-destructive testing and evaluation.