{"id":355,"date":"2024-05-24T13:50:21","date_gmt":"2024-05-24T12:50:21","guid":{"rendered":"https:\/\/eslsrv24.epfl.ch\/kid-ppg\/?page_id=355"},"modified":"2024-05-28T14:06:08","modified_gmt":"2024-05-28T13:06:08","slug":"the-kid-ppg-framework","status":"publish","type":"page","link":"https:\/\/eslweb.epfl.ch\/kid-ppg\/the-kid-ppg-framework\/","title":{"rendered":"The KID-PPG framework"},"content":{"rendered":"\n

At the heart of KID-PPG lie three key insights on PPG, Heart Rate and motion artifacts:<\/p>\n\n\n\n

    \n
  1. Motion artifact removal needs to be explicitely defined as a source separation task.<\/li>\n\n\n\n
  2. In certain cases the PPG might be so corrupted by motion that it is infeasible to recover the heart rate from it – even if we have access to acceleration information through accelerometer sensors.<\/li>\n\n\n\n
  3. The Blood Volume Pulse has specific characteristics. For example we saw in the introduction that a clean PPG signal is composed of the 2 \u2022 HeartRate<\/em><\/strong> components.<\/li>\n<\/ol>\n\n\n\n

    We have designed three mechanisms to integrate this prior-knowledge<\/strong> into KID-PPG:<\/p>\n\n\n\n

      \n
    1. Explicit Source Separation<\/li>\n\n\n\n
    2. Probabilistic Inference and<\/li>\n\n\n\n
    3. Guided Training<\/li>\n<\/ol>\n\n\n\n

      We detail these three mechanisms below, providing also details on how our kid-ppg python package<\/a> can be used for direct heart rate estimation using these three key insights.<\/p>\n\n\n\n

      Our python kid-ppg package is the first ever PPG-based heart rate extraction deep learning model with publicly available pretrained weights. You can easily install it and start analyzing PPG and acceleration data by simply downloading it through pip:<\/p>\n\n\n

      \npip install kid-ppg\n<\/pre><\/div>\n\n\n
      1. Explicit Source Separation<\/strong><\/p>\n\n\n\n
      As we’ve seen, external motions can significantly damage the heart component in the PPG signal. In the case of periodic motions we can get some hints about the motion artifacts by observing the acceleration signals. In this case the 3D acceleration signal acts as a motion reference<\/strong> which we utilize to cancel the motion artifacts from the PPG.<\/p>\n\n\n\n
      Unlike the SoA Deep Learning PPG models, which perform arbitrary data fusion, we propose an explicit source separation task. This way we enable the model to only focus on the most relevant parts of the signal – the Blood Volume Pulse.<\/p>\n\n\n\n
      We chose to implement this source separation task as a linear adaptive filter<\/a>. kid-ppg<\/code> provides the AdaptiveFilteringModel<\/code> for linear adaptive filtering. Calling an AdaptiveFilteringModel<\/code> model first trains the model adaptively based on the input PPG and accelerations. Then, it performs filtering based on the learned weights.<\/p>\n\n\n
      \nimport tensorflow as tf\nfrom kid_ppg.preprocessing import sample_wise_z_score_normalization, sample_wise_z_score_denormalization\nfrom kid_ppg.adaptive_linear_model import AdaptiveFilteringModel\n\n# We define an optimizer which we will use to converge\n# the adaptive model.\nsgd = tf.keras.optimizers.legacy.SGD(learning_rate = 1e-7,\n                                            momentum = 1e-2,)\n# We train the model for 1000 epochs.\nn_epochs = 1000\nmodel = AdaptiveFilteringModel(local_optimizer = sgd,\n                               num_epochs_self_train = n_epochs)\n\n# We simultaneously perform model training and\n# filtering. X_input contains the PPG signals in\n# its first channel and 3 acceleration signals\n# in the rest 3 channels. Overall, X_input size\n# is [N_samples, 256, 4].\nPPG_filtered = model(X_input[..., None]).numpy()\n<\/pre><\/div>\n\n\n
      The resulting PPG_filtered<\/code> contains the the PPG signal after removing the motion artifacts. The results of our exploration showcase how this explicit source separation<\/strong> can significantly boost the performance of multiple deep learning models.

      <\/p>\n\n\n\n
      \n      <\/object>\n<\/div>\n      
      \n\n        
      \n