Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Dr Gusztav Belteki

Dr Gusztav Belteki

Slides presented on the 30th Pydata London Meetup

Gusztav Belteki

January 10, 2017
Tweet

More Decks by Gusztav Belteki

Other Decks in Research

Transcript

  1. Analyzing neonatal ven.lator data with Python Gusztav Belteki Giles Weaver,

    Ian Ozsvald Pydata – 30th Meetup, London, 10/07/2017
  2. Why do some newborn babies require mechanical ven.la.on ?¶ • 

    Prematurity: lung, muscles and brain are too immature to support adequate gas exchange •  Full-term babies may require intensive care (e.g. infec;on, aJer an opera;on, birth depression etc.) •  We ven;late >1500 “ven;lator days” yearly •  Ven;la;on is also an important part of adult intensive care
  3. Mechanical ven.la.on is always a complex physical process due to

    interac.on between the ven.lator and the pa.ent Pa.ent not breathing Pa.ent breathing spontaneously
  4. What are the benefits of new ven0lators? •  Ven;lator-pa;ent interac;on

    has improved significantly: 1.  Synchroniza.on with the pa;ent's own breathing effort 2.  Aiming for a set volume of gas (volume targe.ng) rather than using the same infla;ng pressure
  5. What are the problems with the new ven0lators?¶ •  They

    are "too complex" for everyday clinical work •  Most data the ven;lators are providing are not considered during clinical decision making •  Ven;lator data are not rou;nely archived at a high sampling rate and not analyzed retrospec;vely
  6. Aims •  The collect ven;lator data at high sampling rate

    and analyse them computa;onally •  To provide the clinician with SIMPLE indicators of ven;la;on and ven;lator-pa;ent interac;on •  Eventually to move closer to automa;on of mechanical ven;la;on
  7. Data Collec.on •  Downloaded ~160 days of ven;lator data from

    59 ven;lated neonates •  Most con;nuous recordings are >24 hours, usually 2-4 days •  Data are retrieved as csv files •  Generates approximately 650 Mbyte data / 24 hours of ven;la;on
  8. CPU times: user 46 s, sys: 6.3 s, total: 52.3

    s Wall time: 54 s <class 'pandas.core.frame.DataFrame'> DatetimeIndex: 16379346 entries, 
 2016-06-24 16:57:54.203000 to 
 2016-06-26 14:40:29.289000 Data columns (total 6 columns): ..... memory usage: 874.8 MB Impor.ng “fast” data: >16 million rows imported over 54 seconds !
  9. Flow (L/min) Pressure (mBar) Ar;ficial lung Sedated pa;ent Breathing pa;ent

    fig = sns.jointplot(x="paw", y="flow", data=fast_2a, s = 5) Two-dimensional scaberplots using the seaborn library 3 hours of ven;la;on (~1,000,000 data points)
  10. Acknowledgments:¶ •  Professor Colin Morley •  Draeger Medical¶ •  All

    the doctors and nurses of Cambridge Neonatal Intensive Care Unit