Model-Free Machine Learning in Biomedicine: Feasibility Study in Type 1 Diabetes

被引:39
作者
Daskalaki, Elena [1 ]
Diem, Peter [2 ]
Mougiakakou, Stavroula G. [1 ,2 ]
机构
[1] Univ Bern, ARTORG Ctr Biomed Engn Res, Diabet Technol Res Grp, Murtenstr 50, CH-3008 Bern, Switzerland
[2] Bern Univ Hosp Inselspital, Div Endocrinol Diabet & Clin Nutr, CH-3010 Bern, Switzerland
来源
PLOS ONE | 2016年 / 11卷 / 07期
关键词
ARTIFICIAL PANCREAS SYSTEMS; ACTOR-CRITIC ALGORITHMS; CLOSED-LOOP CONTROL; RANDOMIZED CROSSOVER; GLUCOSE REGULATION; ADULTS; CHALLENGES; PREDICTION;
D O I
10.1371/journal.pone.0158722
中图分类号
O [数理科学和化学]; P [天文学、地球科学]; Q [生物科学]; N [自然科学总论];
学科分类号
07 ; 0710 ; 09 ;
摘要
Although reinforcement learning (RL) is suitable for highly uncertain systems, the applicability of this class of algorithms to medical treatment may be limited by the patient variability which dictates individualised tuning for their usually multiple algorithmic parameters. This study explores the feasibility of RL in the framework of artificial pancreas development for type 1 diabetes (T1D). In this approach, an Actor-Critic (AC) learning algorithm is designed and developed for the optimisation of insulin infusion for personalised glucose regulation. AC optimises the daily basal insulin rate and insulin: carbohydrate ratio for each patient, on the basis of his/her measured glucose profile. Automatic, personalised tuning of AC is based on the estimation of information transfer (IT) from insulin to glucose signals. Insulin-to-glucose IT is linked to patient-specific characteristics related to total daily insulin needs and insulin sensitivity (SI). The AC algorithm is evaluated using an FDA-accepted T1D simulator on a large patient database under a complex meal protocol, meal uncertainty and diurnal SI variation. The results showed that 95.66% of time was spent in normoglycaemia in the presence of meal uncertainty and 93.02% when meal uncertainty and SI variation were simultaneously considered. The time spent in hypoglycaemia was 0.27% in both cases. The novel tuning method reduced the risk of severe hypoglycaemia, especially in patients with low SI.
引用
收藏
页数:20
相关论文
共 57 条
  • [11] Cameron Fraser, 2011, J Diabetes Sci Technol, V5, P368
  • [12] Dynamic Reconfiguration of Shipboard Power Systems Using Reinforcement Learning
    Das, Sanjoy
    Bose, Sayak
    Pal, Siddharth
    Schulz, Noel N.
    Scoglio, Caterina M.
    Natarajan, Bala
    [J]. IEEE TRANSACTIONS ON POWER SYSTEMS, 2013, 28 (02) : 669 - 676
  • [13] Daskalaki E, 2013, P 35 IEEE ENG MED BI
  • [14] An Actor-Critic based controller for glucose regulation in type 1 diabetes
    Daskalaki, Elena
    Diem, Peter
    Mougiakakou, Stavroula G.
    [J]. COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE, 2013, 109 (02) : 116 - 125
  • [15] Real-Time Adaptive Models for the Personalized Prediction of Glycemic Profile in Type 1 Diabetes Patients
    Daskalaki, Elena
    Prountzou, Aikaterini
    Diem, Peter
    Mougiakakou, Stavroula G.
    [J]. DIABETES TECHNOLOGY & THERAPEUTICS, 2012, 14 (02) : 168 - 174
  • [16] Closed-Loop Insulin Therapy Improves Glycemic Control in Children Aged <7 Years A randomized controlled trial
    Dauber, Andrew
    Corcia, Liat
    Safer, Jason
    Agus, Michael S. D.
    Einis, Sara
    Steil, Garry M.
    [J]. DIABETES CARE, 2013, 36 (02) : 222 - 227
  • [17] On-line policy learning and adaptation for real-time personalization of an artificial pancreas
    De Paula, Mariano
    Acosta, Gerardo G.
    Martinez, Ernesto C.
    [J]. EXPERT SYSTEMS WITH APPLICATIONS, 2015, 42 (04) : 2234 - 2255
  • [18] Donsa K, INSMART HLTH 2015, P237
  • [19] Closed- Loop Artificial Pancreas Systems: Engineering the Algorithms
    Doyle, Francis J., III
    Huyett, Lauren M.
    Lee, Joon Bok
    Zisser, Howard C.
    Dassau, Eyal
    [J]. DIABETES CARE, 2014, 37 (05) : 1191 - 1197
  • [20] Robust Automated Detection of Microstructural White Matter Degeneration in Alzheimer's Disease Using Machine Learning Classification of Multicenter DTI Data
    Dyrba, Martin
    Ewers, Michael
    Wegrzyn, Martin
    Kilimann, Ingo
    Plant, Claudia
    Oswald, Annahita
    Meindl, Thomas
    Pievani, Michela
    Bokde, Arun L. W.
    Fellgiebel, Andreas
    Filippi, Massimo
    Hampel, Harald
    Kloeppel, Stefan
    Hauenstein, Karlheinz
    Kirste, Thomas
    Teipel, Stefan J.
    [J]. PLOS ONE, 2013, 8 (05):