Explainability Metrics of Deep Convolutional Networks for Photoplethysmography Quality Assessment

Abstract

Photoplethysmography (PPG) is a noninvasive way to monitor various aspects of the circulatory system, and is becoming more and more widespread in biomedical processing. Recently, deep learning methods for analyzing PPG have also become prevalent, achieving state of the art results on heart rate estimation, atrial fibrillation detection, and motion artifact identification. Consequently, a need for interpretable deep learning has arisen within the field of biomedical signal processing. In this paper, we pioneer novel explanatory metrics which leverage domain-expert knowledge to validate a deep learning model. We visualize model attention over a whole testset using saliency methods and compare it to human expert annotations. Congruence, our first metric, measures the proportion of model attention within expert-annotated regions. Our second metric, Annotation Classification, measures how much of the expert annotations our deep learning model pays attention to. Finally, we apply our metrics to compare between a signal based model and an image based model for PPG signal quality classification. Both models are deep convolutional networks based on the ResNet architectures. We show that our signal-based one dimensional model acts in a more explainable manner than our image based model; on average 50.78% of the one dimensional model's attention are within expert annotations, whereas 36.03% of the two dimensional model's attention are within expert annotations. Similarly, when thresholding the one dimensional model attention, one can more accurately predict if each pixel of the PPG is annotated as artifactual by an expert. Through this testcase, we demonstrate how our metrics can provide a quantitative and dataset-wide analysis of how explainable the model is.

Beat-Based PPG-ABP Cleaning Technique for Blood Pressure Estimation

“…However, this simple labeling ignores the minimum level of morphological quality completely. Both Machine learning and deep-learning techniques [17,18,22,42,43] are employed for data cleaning or classifying signals as acceptable or anomalous. However, most of these approaches rely on manual data labeling based on experts' annotation.…”

Design of an intrinsically scalable model for medical disease identification using hybrid swarm intelligence with augmented transfer learning

“…These models assist in estimation of highly variant features, thereby improving overall classification performance, without compromising on delay & scalability performance. Performance of these models must be validated under different datasets, and can be extended via use of Dense Convolutional Neural Network [17], fuzzy logic-based classification [18], Deep Multiple instance Learning via Multiple Modalities [19], ensemble classification models [20], semi supervised Multitask Learning (SSMTL) [21], and Variance Maximized Deep Networks [22], which assist in improving feature variance for better signal representation under different datasets. Specialized approaches that utilize Deep Learning Models for Keratoconus and Sub-Clinical Keratoconus Detection [23], Multivariate Analysis using CNNs & Decision Trees [24], Learning from Label Fuzzy Proportions (LLFP) [25], and Gaussian mixture models (GMM) with sensor CNN (SCNN) [26] are discussed which assist in improving correlative feature mapping between clinical & test features in order to improve overall classification performance.…”

Applications of Explainable AI for 6G: Technical Aspects, Use Cases, and Research Challenges

“…The QoT contains a physical and emotional trust, representing the objective and subjective assessment of explainability, respectively. For analysing photoplethysmography using deep learning methods, Zhang et al [267] proposed two metrics (i.e., congruence, annotations classifications) to measure the quality of XAI explanations compared to the human experts' explanations. Similarly, Kaur et al [268] proposed another metric called "Trustworthy Explainability Acceptance" that measures the Euclidean distance between XAI explanations and domain experts' reasonings in predicting Ductal Carcinoma in Situ (DCIS) recurrence using AI.…”