534
B. Farahani et al.
In short, timed bio-signals are monitored by sensors in the IoT e-Health device
and those signals are transmitted to fog nodes. The fog nodes then sort, process,
extract, and compress the data before the processed, timestamped data is sent to the
cloud through a secure connection. Next, the data is converted to a sparse distributed
representation (SDR). The SDRs are then put through the HTM model, which
attempts to mimic neocortex brain activities by learning temporal SDR patterns
constantly. Over time, the HTM creates a complex, adaptive model that is able
to predict the next incoming SDR sequence. If the next actual sequence does not
match the predicted sequence, an anomaly alert is triggered. Reducing data variance
induced false-positive results requires that HTM calculate a time-varying average of
the error that is compared with the distribution of all errors. This comparison enables
HTM to forecast the accuracy level of the predicted anomaly or alert. The outputs
generated by the risk analysis and warnings can be viewed via a dashboard and
sent to patients and medical professionals tasked with providing the best possible
treatment.
11.5.8 Population Health Management
IBM has estimated that available medical data will double every 73 days by 2020
[17]. Big Data analytics can be used to help comprehend the data and extract deep
insights in order to improve medical outcomes and reduce costs through more
customized care plans and early intervention. IBM has also reported that the Medical
Center of Columbia University used Big Data analytics to review medical data
from patients suffering from strokes in order to predict serious complications up to
48 hours before traditional models. The Rizzoli Orthopedic Institute also used Big
Data analytics to better understand the clinical differences in families with genetic
bone diseases, resulting in a 30% decrease in yearly hospitalizations. Similarly, the
Hospital for Sick Children reviewed diverse Big Data analysis methods to monitor
and process patient vital signs and predict infections up to 24 hours faster than
traditional methods. Additionally, machine learning methods have been considered
for automatically monitoring psycho-physiological stress from bio-signals including
skin conductance and an accelerometer. Generally, Big Data analytics is useful in
answering the important questions listed below:
• How can alarm-based screenings be used to predict which patients are at the most
risk?
• Which treatments and plans of care generate the best patient outcomes?
• How can bio-signals and medical data be processed to accurately predict
complications?
• How can an automatic diagnostic system use classification techniques to help
patients even if a medical professional is not present?
• How can a recommendation system using collaborative filtering systems be
created to suggest the most beneficial treatment or HCC gene-based treatment
based on similar measures between patients or treatments?
B. Farahani et al.
In short, timed bio-signals are monitored by sensors in the IoT e-Health device
and those signals are transmitted to fog nodes. The fog nodes then sort, process,
extract, and compress the data before the processed, timestamped data is sent to the
cloud through a secure connection. Next, the data is converted to a sparse distributed
representation (SDR). The SDRs are then put through the HTM model, which
attempts to mimic neocortex brain activities by learning temporal SDR patterns
constantly. Over time, the HTM creates a complex, adaptive model that is able
to predict the next incoming SDR sequence. If the next actual sequence does not
match the predicted sequence, an anomaly alert is triggered. Reducing data variance
induced false-positive results requires that HTM calculate a time-varying average of
the error that is compared with the distribution of all errors. This comparison enables
HTM to forecast the accuracy level of the predicted anomaly or alert. The outputs
generated by the risk analysis and warnings can be viewed via a dashboard and
sent to patients and medical professionals tasked with providing the best possible
treatment.
11.5.8 Population Health Management
IBM has estimated that available medical data will double every 73 days by 2020
[17]. Big Data analytics can be used to help comprehend the data and extract deep
insights in order to improve medical outcomes and reduce costs through more
customized care plans and early intervention. IBM has also reported that the Medical
Center of Columbia University used Big Data analytics to review medical data
from patients suffering from strokes in order to predict serious complications up to
48 hours before traditional models. The Rizzoli Orthopedic Institute also used Big
Data analytics to better understand the clinical differences in families with genetic
bone diseases, resulting in a 30% decrease in yearly hospitalizations. Similarly, the
Hospital for Sick Children reviewed diverse Big Data analysis methods to monitor
and process patient vital signs and predict infections up to 24 hours faster than
traditional methods. Additionally, machine learning methods have been considered
for automatically monitoring psycho-physiological stress from bio-signals including
skin conductance and an accelerometer. Generally, Big Data analytics is useful in
answering the important questions listed below:
• How can alarm-based screenings be used to predict which patients are at the most
risk?
• Which treatments and plans of care generate the best patient outcomes?
• How can bio-signals and medical data be processed to accurately predict
complications?
• How can an automatic diagnostic system use classification techniques to help
patients even if a medical professional is not present?
• How can a recommendation system using collaborative filtering systems be
created to suggest the most beneficial treatment or HCC gene-based treatment
based on similar measures between patients or treatments?