ECG Waveform Extraction From Distorted Images Using DNN Segmentation

The patent replaces traditional mechanical thresholding methods with a deep neural network-based automated extraction system. The DNN model learns complex patterns from training data and automatically distinguishes ECG signals from grid lines, handling noise and distortions that confuse simple thresholding methods. This substitution of manual/algorithmic thresholds with intelligent neural networks resolves the contradiction between simplicity and precision.

The patent transforms the extraction approach by changing from fixed threshold parameters to adaptive neural network parameters. The DNN model adjusts its internal parameters based on input image characteristics, allowing it to handle varying paper qualities, lighting conditions, and distortions. This parameter transformation enables the system to maintain high signal quality across diverse conditions while remaining computationally efficient.

The patent applies preliminary action by pre-training the deep neural network on a comprehensive dataset of ECG images with various distortions and qualities. Once trained, the model can quickly extract signals from new images without requiring complex real-time computations. The heavy computational burden is shifted to the offline training phase, allowing fast inference during actual use, thus resolving the complexity-accuracy contradiction.

The patent uses a pre-trained neural network model that has copied and learned from numerous examples during training. This pre-trained model serves as a reusable computational template that can process new ECG images efficiently. By copying the learned patterns from training data, the system achieves high accuracy without requiring complex adaptive computations for each new image, reducing real-time computational complexity.