Microwave Dielectric Imaging Using ML for Reliable Disease Detection

The patent replaces conventional nonlinear mechanical iterative inversion algorithms with a machine learning-based system. A neural network is trained on simulated microwave scattering data to directly predict dielectric profiles, eliminating the need for complex iterative mathematical inversion processes. This substitution transforms the unreliable nonlinear reconstruction problem into a reliable pattern recognition task, achieving both improved reliability and reduced operational complexity.

The patent applies preliminary action by pre-training a machine learning model on extensive simulated scattering data before actual measurement. The neural network learns the mapping between scattering patterns and dielectric profiles in advance during the training phase. During actual measurement, the pre-trained model can quickly predict dielectric profiles from limited scattering data without requiring extensive additional measurements, thus reducing acquisition time while maintaining precision.

The patent uses simulated scattering data as a copy or surrogate for actual measured data during the training phase. By training the machine learning model on extensively simulated scattering patterns that represent various dielectric configurations, the system learns from this virtual data copy without requiring a physically complex antenna array. This allows the model to be trained on complete scattering information while the actual measurement system remains simpler.

The patent replaces complex conventional MWT reconstruction algorithms with a trained machine learning model that directly predicts dielectric properties from scattering measurements. This substitution transforms the operationally difficult iterative inversion process into a simple forward prediction task, maintaining measurement precision while dramatically improving ease of operation for routine clinical use.