Key result
Multimodal 1D-to-2D CNN classifier achieves ~92% disease classification accuracy without manual preprocessing.
Why the study?
Conventional 1D pulse disease recognition methods face challenges of resemblance, ambiguity, low accuracy, and instability in identifying multiple diseases.
Transforming 1D pulse signals into 2D representations for CNN classification achieves up to 92% accuracy, offering a promising approach for computer-aided pulse diagnosis.
May aid computer-aided pulse diagnosis; leaves open prospective clinical validation before practice adoption.
Pulse signals, vital indicators of intrinsic health information, hold diverse applications in computer-aided pulse diagnosis in traditional Chinese medicine. However, conventional one-dimensional pulse disease recognition methods face challenges of resemblance, ambiguity, low accuracy, and instability in identifying multiple diseases. To address these issues, this study presents a novel approach that uses multiple methods to transform one-dimensional signals into two-dimensional representations. Initially, original signals are segmented into periods, treating single-period signals as samples. Then, employing short-time Fourier transform, Gram angle field, Markov variation field, and recurrence map, 2D matrices, and corresponding image representations are derived across time, frequency, and time-frequency domains. An integrated model is established to classify multiple images as features. Experiments utilize a pulse dataset from 211 hospitals for training and testing. Results reveal that devoid of manual preprocessing, signal upscaling enhances disease classification accuracy by capturing both temporal and periodic/non-periodic signal information. This approach achieves an average accuracy of up to 92%. The synergy of signal upscaling and an integrated CNN classifier demonstrates promising potential in pulse disease recognition.
No takes yet. Share an insight, caveat, or question.
Fan et al. (2023) studied Pulse disease recognition. Multimodal 1D-to-2D signal transformation and integrated CNN classifier was evaluated on Disease classification accuracy. A multimodal 1D-to-2D signal transformation approach using an integrated CNN classifier achieved an average disease classification accuracy of up to 92% without manual preprocessing.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: