Multi-Hour Blood Glucose Prediction in Type 1 Diabetes: A Patient-Specific Approach Using Shallow Neural Network Models

Kushner, Taisa; Breton, Marc D.; Sankaranarayanan, Sriram

doi:10.1089/dia.2020.0061

Cited by 29 publications

(18 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Few studies focused on prediction of continuous outcomes (e.g. glucose levels [26, 27]) and forecasts on the population level (e.g. COVID-19 trend [28] and dengue fever outbreaks [29]).…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Transfer learning for non-image data in clinical research: a scoping review

Ebbehøj

Thunbo

Andersen

et al. 2021

Preprint

View full text Add to dashboard Cite

Background Transfer learning is a form of machine learning where a pre-trained model trained on a specific task is reused as a starting point and tailored to another task in a different dataset. While transfer learning has garnered considerable attention in medical image analysis, its use for clinical non-image data is not well studied. Therefore, the objective of this scoping review was to explore the use of transfer learning for non-image data in the clinical literature. Methods and Findings We systematically searched medical databases (PubMed, EMBASE, CINAHL) for peer-reviewed clinical studies that used transfer learning on human non-image data. We included 83 studies in the review. More than half of the studies (63%) were published within 12 months of the search. Transfer learning was most often applied to time series data (61%), followed by tabular data (18%), audio (12%) and text (8%). Thirty-three (40%) studies applied an image-based model to non-image data after transforming data into images (e.g. spectrograms). Twenty-nine (35%) studies did not have any authors with a health-related affiliation. Many studies used publicly available datasets (66%) and models (49%), but fewer shared their code (27%). Conclusions In this scoping review, we have described current trends in the use of transfer learning for non-image data in the clinical literature. We found that the use of transfer learning has grown rapidly within the last few years. We have identified studies and demonstrated the potential of transfer learning in clinical research in a wide range of medical specialties. More interdisciplinary collaborations and the wider adaption of reproducible research principles are needed to increase the impact of transfer learning in clinical research.

show abstract

“…Few studies focused on prediction of continuous outcomes (e.g. glucose levels [26, 27]) and forecasts on the population level (e.g. COVID-19 trend [28] and dengue fever outbreaks [29]).…”

Section: Discussionmentioning

confidence: 99%

“…Moreover, transfer learning was used for prediction of glucose levels [26,27], estimation of Parkinson's disease severity [71], detection of cognitive impairment [33] and schizophrenia [52], and forecasting of infectious disease trends [28] and outbreaks [29], among other applications [72][73][74][75][76][77][78][79][80][81][82].…”

Section: Time Seriesmentioning

confidence: 99%

Transfer learning for non-image data in clinical research: a scoping review

Ebbehøj

Thunbo

Andersen

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…In order to allow for the natural variations in the BGL regulation system characteristics, prediction models are usually trained and evaluated patient-by-patient [ 7 , 8 , 9 , 10 , 11 ]. An alternative approach is applying the transfer learning principle, which means that a prediction model developed for a patient is used as the startup model in the training for another patient [ 12 , 13 , 14 , 15 ].…”

Section: Introductionmentioning

confidence: 99%

“…The published studies used a wide variety of technologies for the prediction, ranging from physiological BGL regulation models through support vector regression models to neuro-fuzzy inferencing, but most of all, neural networks (NNs) of several kinds. The fundamental problem of using BGL regulation models is that, to produce reliable results, the model must be sufficiently complex, but, the more complex the model, the more parameters it has, and thus the harder it is to personalize [ 14 , 17 , 18 ]. The other methods use no BGL regulation model, only a sufficiently long BGL/lifestyle log, and artificial intelligence to learn the complex relationships between BGL and the elements of lifestyle (diet, insulin, and even other factors such as stress or physical activity).…”

Section: Introductionmentioning

confidence: 99%

“…If the patient wears a CGM all the time and follows a regular daily meal schedule, the model may be based purely on the CGM and, optionally, insulin data, without any dietary information. The latest result of this kind was presented by Kushner et al, who developed a feed-forward NN (FNN)-based model for 24 patients with good results on the mid-term PHs (RMSE 120 min: 2.11, 180 min: 2.2 mmol/L) [ 14 ]. A similar prediction accuracy (120 min: 2.08 mmol/L) was reached by Zarkogianni et al, earlier in 2015 with an adaptive neuro-fuzzy system, though the 10-person study also used information on physical activity [ 9 ].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Improved Methods for Mid-Term Blood Glucose Level Prediction Using Dietary and Insulin Logs

2021

View full text Add to dashboard Cite

Background and Objectives: The daily lifestyle management of diabetes requires accurate predictions of the blood glucose level between meals. The objective of this study was to improve the accuracy achieved by previous work, especially on the mid-term, i.e., 120 to 180 min prediction horizons, for insulin-dependent patients. Materials and Methods: An absorption model-based method is proposed to train an artificial neural network with the bolus and basal insulin dosing and timing, the baseline blood glucose level, the maximal glucose infusion rate, and the total carbohydrate content as parameters. The approach was implemented in various algorithmic setups, and it was validated on data from a small-scale clinical trial with continuous glucose monitoring. Results: Root mean square error results for the mid-term horizons are 1.72 mmol/L (120 min) and 1.95 mmol/L (180 min). The accuracy of the proposed model measured on the clinical data is better than the accuracy reported by any other currently available and comparable models. Conclusions: A relatively short (ca. two weeks) training sample of a continuous glucose monitor and dietary/insulin log is sufficient to provide accurate predictions. For the outpatient application in practice, a hybrid model is proposed that combines the present mid-term method with the authors’ previous work for short-term predictions.

show abstract