PAFC signal identification method and device based on deep learning

Through deep learning-based methods, the features of PAFC signals are extracted and processed, and the ResNet50 model and Logistic regression model are used for identification, which solves the problem of difficult identification of artifacts and false positive signals in the prior art, and achieves efficient identification of melanoma signals.

CN119961625APending Publication Date: 2025-05-09GUANGYU BIOMEDICAL TECHNOLOGY (SUZHOU) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510027892.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-01-08
Publication Date
2025-05-09

AI Technical Summary

Technical Problem

In existing PAFC technology, artifacts caused by jitter and external interference, as well as false positive photoacoustic signals, are difficult to identify, affecting the early screening and treatment of melanoma.

Method used

Using a deep learning-based method, the photoacoustic signal is obtained for baseline correction and normalization preprocessing, and features such as peak value, Pearson coefficient, gradient feature are extracted. Combined with fast Fourier transform and Hilbert transform, the photoacoustic signal feature fingerprint map is generated, and the trained ResNet50 model is input for identification. The threshold is determined using the Logistic regression model to realize the identification of the PAFC target signal.

Benefits of technology

It realizes fast, efficient and accurate identification of melanoma signals and noise artifact signals, reduces labeling costs, and improves automated data processing capabilities and classification performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119961625A_ABST
    Figure CN119961625A_ABST
Patent Text Reader

Abstract

The invention provides a PAFC signal identification method and device based on deep learning. The method comprises the following steps: acquiring a to-be-identified signal; preprocessing the to-be-identified signal and extracting a feature vector; performing fast Fourier transform and Hilbert transform on the preprocessed to-be-identified signal to obtain frequency domain waveform data and an analysis signal; fusing to obtain a photoacoustic signal characteristic fingerprint spectrum; inputting the photoacoustic signal feature fingerprint spectrum into a trained transfer learning model, and generating a sequence which changes along with time in a time range corresponding to the preprocessed to-be-identified signal and an identification result of each element in the sequence; and counting and timing the recognition result by a counter, and taking the preprocessed to-be-recognized signal of which the number of the circulating tumor cells and the circulating tumor cell clusters is greater than or equal to a second threshold value as a PAFC target signal. The method has the advantages of low labeling cost, high-speed automatic data processing capacity and classification superior to that of a traditional method.
Need to check novelty before this filing date? Find Prior Art