MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202641048483 A) filed by Seshadri Rao Gudlavalleru Engineering College; Balakrishna Tilakachuri; Krishna Sai Saran Veeramallu; Hemachand Ravulapalli; Anusha Rani Devarakonda; and Kusal Sai Kiran Sajja, Gudlavalleru, Andhra Pradesh, on April 16, for 'enhanced ecg arrhythmia prediction using multi-lead signals and hybrid ai models.'
Inventor(s) include Seshadri Rao Gudlavalleru Engineering College; Balakrishna Tilakachuri; Krishna Sai Saran Veeramallu; Hemachand Ravulapalli; Anusha Rani Devarakonda; and Kusal Sai Kiran Sajja.
The application for the patent was published on May 1, under issue no. 18/2026.
According to the abstract released by the Intellectual Property India: "Early detection of cardiovascular abnormalities is critical for lowering mortality risk and enabling prompt clinical management. This work presents a noise-resilient hybrid ECG classification framework that combines clinical benchmark datasets (MIT-BIH Arrhythmia Database and PTB-XL) with portable six-lead recordings obtained from KardiaMobile 6L devices. ECG traces from portable PDF and image reports are automatically digitized into standardized waveforms using an image-processing pipeline with signal-quality-driven lead selection, whereas multi-lead clinical signals are compacted through principal component analysis (PCA). To ensure uniformity across heterogeneous sources, all recordings are resampled to 100 Hz and segmented into fixed-length windows prior to preprocessing. The proposed denoising pipeline sequentially applies Butterworth band-pass filtering, power-line notch suppression, discrete wavelet denoising, and per-record normalization. For classification, the framework integrates both deep and conventional machine learning models. A one-dimensional CNN captures temporal ECG morphology, while a two-dimensional CNN analyzes STFT-derived time-frequency representations. In parallel, handcrafted clinical features are processed using support vector machine (SVM), random forest (RF), XGBoost, and k-nearest neighbors (KNN) classifiers. Model predictions are combined through a validation-guided weighted probability ensemble to enhance robustness and decision stability. The system performs five-class rhythm identification: Normal Sinus Rhythm, Atrial Fibrillation, Bradycardia, Tachycardia, and Ventricular Arrhythmias. Patient-wise data partitioning is applied to prevent subject information leakage and the framework is evaluated on both intra-dataset digital test data and on Kardia data from an external source. Experiments show high accuracy better macro-F1 performance and well calibrated confidence scores indicating this can be applied to real-time and telecardiology ECG monitoring."
Disclaimer: Curated by HT Syndication.
