AI-Assisted Annotator Using Reinforcement Learning
- PDF / 908,124 Bytes
- 8 Pages / 595.276 x 790.866 pts Page_size
- 81 Downloads / 209 Views
ORIGINAL RESEARCH
AI‑Assisted Annotator Using Reinforcement Learning V. Ratna Saripalli 1 · Dibyajyoti Pati 1 · Michael Potter 1 · Gopal Avinash 1 · Charles W. Anderson2 Received: 25 July 2020 / Accepted: 28 September 2020 © Springer Nature Singapore Pte Ltd 2020
Abstract Machine learning in the healthcare domain is often hindered by data which are both noisy and lacking reliable ground truth labeling. Moreover, the cost of cleaning and annotating this data is significant since, unlike other data domains, medical data annotation requires the work of skilled medical professionals. In this work, we introduced the use of reinforcement learning to mimic the decision-making process of annotators for medical events allowing automation of annotation and labeling. Our reinforcement agent learns to annotate health monitor alarm data based on annotations done by an expert. We demonstrate the efficacy of our implementation on ICU critical alarm data sets. We evaluate our algorithm against standard supervised machine learning and deep learning methods. Compared to SVM and LSTM methods, our method achieves high sensitivity that is critical for alarm data; exhibits better generalization across mixed downsampling; and preserves comparable model performance. Keywords False alarms · Reinforcement learning · Annotation
Introduction Healthcare costs have increased significantly worldwide over the past decades, and especially within the United States where they now account for nearly 18% of the GDP. As such, it is imperative to find ways to lower systemic costs while not compromising quality of care. In particular, Operating Room (OR) costs represent a significant fraction of a hospital’s expenditures. Many sequential decision-making steps are involved in the day-to-day function of an OR. Common examples include deciding when to transfer patients to post-anesthesia care units, scheduling staff for care units, * V. Ratna Saripalli [email protected] Dibyajyoti Pati [email protected] Michael Potter [email protected] Gopal Avinash [email protected] Charles W. Anderson [email protected] 1
GE Healthcare, 2623 Camino Ramon, San Ramon, CA 94583, USA
Colorado State University, Fort Collins, CO 80523, USA
2
determining surgery end, estimating emergence phase, estimating time to extubate, and setting critical event alarms. These problems present attractive targets for novel machine learning applications which, leveraging an ever-growing corpus of medical data, may be able to significantly streamline hospital logistic operations. Data are the fundamental currency for solving many healthcare problems using computational methods. While volumes of medical data are increasingly being made available, these datasets bring their own unique challenges. Medical data are plagued with concerns ranging from data privacy and ground truth availability, to domain issues such as sparsity and heterogeneity, to quality issues such as missing data and noise [1–4]. Moreover, obtaining trustworthy annotations of medica
Data Loading...
