Feature-based diagnostic distortion measure for unsupervised self-guided biomedical signal compressors

In this work, the advantages of coupling biomedical signal compressors with clinical feature-based distortion measures are demonstrated. Such a coupling allow biomedical signal compressors to self-establish hard limits with regards to choices surrounding compression ratios, or 'quality settings', a compressor can safely choose from to guarantee that features of clinical significance are protected so that their reconstruction remains clinically relevant. This coupling allows biomedical signal compressors to operate in an unsupervised manner, since it is demonstrated that establishing hard limits that are applied equally to all signals does not allow one to maximize and/or strike a balance between compression ratio and signal fidelity. Such mechanisms can be employed in communication architectures in wearable body area sensor networks (BASNs) for emerging Internet of Things (IoT) applications for autonomous tasks. While feature-based distortion measures such as the Clinical Distortion Index (CDI), and the Weighted Distortion Measure (WDD) already exist, we demonstrate the viability of our work by proposing a generalizable feature-based distortion measure we call the Diagnostic Distortion Measure (DDM), which offers several benefits that address a few shortcomings present in the CDI and WDD in real-time applications for unsupervised self-guided compressors. Experimental results show successful application of our DDM with ECG signals from the PhysioNet database.