A clinical deep learning framework for continually learning from cardiac signals across diseases, time, modalities, and institutions

Kiyasseh, Dani; Zhu, Tingting; Clifton, David A.

doi:10.1038/s41467-021-24483-0

Cited by 57 publications

(70 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For example, AUC = 0.823 → 0.702 for the cold cut (c) gesture in the USC and St. Antonius settings, respectively. This was expected due to the potential shift in the distribution of data collected across the two medical centres, which has been documented to negatively affect network performance 33 . Potential sources of distribution shift include variability in how surgeons perform the same set of gestures (e.g., different techniques) and the camera recording devices between the surgical robots.…”

Section: Across Medical Centresmentioning

confidence: 99%

Quantification of Robotic Surgeries with Vision-Based Deep Learning

Kiyasseh¹,

Ma²,

Haque³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Surgery is a high-stakes domain where surgeons must navigate critical anatomical structures and actively avoid potential complications while achieving the main task at hand. Such surgical activity has been shown to affect long-term patient outcomes. To better understand this relationship, whose mechanics remain unknown for the majority of surgical procedures, we hypothesize that the core elements of surgery must first be quantified in a reliable, objective, and scalable manner. We believe this is a prerequisite for the provision of surgical feedback and modulation of surgeon performance in pursuit of improved patient outcomes. To holistically quantify surgeries, we propose a unified deep learning framework, entitled Roboformer, which operates exclusively on videos recorded during surgery to independently achieve multiple tasks: surgical phase recognition (the what of surgery), gesture classification and skills assessment (the how of surgery). We validated our framework on four video-based datasets of two commonly-encountered types of steps (dissection and suturing) within minimally-invasive robotic surgeries. We demonstrated that our framework can generalize well to unseen videos, surgeons, medical centres, and surgical procedures. We also found that our framework, which naturally lends itself to explainable findings, identified relevant information when achieving a particular task. These findings are likely to instill surgeons with more confidence in our framework's behaviour, increasing the likelihood of clinical adoption, and thus paving the way for more targeted surgical feedback.

show abstract

Section: Across Medical Centresmentioning

confidence: 99%

Quantification of Robotic Surgeries with Vision-Based Deep Learning

Kiyasseh¹,

Ma²,

Haque³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…It learns a new task at every moment and each task corresponds to one data distribution. Replay-based methods [14,21] retrain the model with old data to consolidate memory and focus on the storage limitation. But in their settings, the old task and new task are clear so that the multi-distribution is fixed.…”

Section: Continuous Classification: Classifying At Every Timementioning

confidence: 99%

“…Meanwhile, as β is the coefficient of loss, if β = 0, the loss are hard to be optimized. Thus, inspired by [14], we introduce a regularization term (β − 1) 2 and initialize β = 1 to penalize it when rapidly decaying toward 0.…”

Section: C3tsmentioning

confidence: 99%

See 1 more Smart Citation

Confidence-Guided Learning Process for Continuous Classification of Time Series

Sun,

Song,

Can

et al. 2022

Preprint

View full text Add to dashboard Cite

In the real world, the class of a time series is usually labeled at the final time, but many applications require to classify time series at every time. e.g. the outcome of a critical patient is only determined at the end, but he should be diagnosed at all times to facilitate timely treatment. Thus, in this paper, we propose a new concept: Continuous Classification of Time Series (CCTS). There are two open problems about CCTS: (1) Data arrangement. Time series is a kind of dynamic data. It evolves multiple distributions over time. The division of multi-distribution will directly affect the classification accuracy; (2) Model training strategy. When a model learns multi-distributed data, it always forgets the old distribution or overfits in the main distribution. Different data learning orders will result in different model performances. We find that the process of model learning multiple distributions can be similar to the process of human learning multiple knowledge. Thus, we propose a novel Confidence-guided method for CCTS to arrange the data and schedule the training, named C 3 TS. It imitates the objective-confidence and the alternating self-confidence of humans in their

show abstract

“…Despite the need for robust continual learning algorithms in clinical settings, applications of such methods to clinical predictive modeling remain scarce 32 . Here we consider a clinically significant problem involving prediction of sepsis in critically ill patients.…”

Section: Introductionmentioning

confidence: 99%

Leveraging clinical data across healthcare institutions for continual learning of predictive risk models

Amrollahi

Shashikumar

Holder

et al. 2022

Sci Rep

View full text Add to dashboard Cite

The inherent flexibility of machine learning-based clinical predictive models to learn from episodes of patient care at a new institution (site-specific training) comes at the cost of performance degradation when applied to external patient cohorts. To exploit the full potential of cross-institutional clinical big data, machine learning systems must gain the ability to transfer their knowledge across institutional boundaries and learn from new episodes of patient care without forgetting previously learned patterns. In this work, we developed a privacy-preserving learning algorithm named WUPERR (Weight Uncertainty Propagation and Episodic Representation Replay) and validated the algorithm in the context of early prediction of sepsis using data from over 104,000 patients across four distinct healthcare systems. We tested the hypothesis, that the proposed continual learning algorithm can maintain higher predictive performance than competing methods on previous cohorts once it has been trained on a new patient cohort. In the sepsis prediction task, after incremental training of a deep learning model across four hospital systems (namely hospitals H-A, H-B, H-C, and H-D), WUPERR maintained the highest positive predictive value across the first three hospitals compared to a baseline transfer learning approach (H-A: 39.27% vs. 31.27%, H-B: 25.34% vs. 22.34%, H-C: 30.33% vs. 28.33%). The proposed approach has the potential to construct more generalizable models that can learn from cross-institutional clinical big data in a privacy-preserving manner.

show abstract

A clinical deep learning framework for continually learning from cardiac signals across diseases, time, modalities, and institutions

Cited by 57 publications

References 13 publications

Quantification of Robotic Surgeries with Vision-Based Deep Learning

Quantification of Robotic Surgeries with Vision-Based Deep Learning

Confidence-Guided Learning Process for Continuous Classification of Time Series

Leveraging clinical data across healthcare institutions for continual learning of predictive risk models

Contact Info

Product

Resources

About