Mid CloudPhy
Mid CloudPhy
Mid CloudPhy
0
IIT Kanpur
Mid Prep
INTRODUCTION
Patient monitoring is a crucial aspect of healthcare, as it allows healthcare professionals
to closely track a patient's vital signs and detect any potential issues before they become
serious. In particular, monitoring a patient's vitals, such as heart rate, blood pressure, and
oxygen levels can provide valuable information about a patient's overall health and
well-being. By closely monitoring vitals, healthcare professionals can quickly identify any
abnormal changes, and take appropriate action to address any concerns. Additionally,
with the increasing use of technology in healthcare, there are now many digital
monitoring systems available that can help to automate the process of tracking vitals,
making it more efficient and accurate. Overall, monitoring vitals is a critical aspect of
providing high-quality care to patients and is essential for ensuring the best possible
patient outcomes. According to the current guidelines, the nurse-to-patient ratio is 1:6,
however, in the real world, the situation is much worse.
PROBLEM STATEMENT
The core problem statement is to extract Heart Rate, SpO2, RR, Systolic Blood Pressure,
Diabolic Blood Pressure, and MAP from the provided images.
The provided images look like the following image:
EVALUATION METRICS
The evaluation metrics include the following:
Accuracy of the extracted information - 60%
Inference time of the entire pipeline - 10%
Novelty - 30%
Brownie points to teams that detect the H.R graph and digitize it.
DATA PROVIDED
Monitor Segmentation Dataset - A dataset containing the segmentation boundaries
for the monitors in the image, a total of 2000 images.
https://drive.google.com/drive/folders/1qCcqEEeTXO9t7dAmG09_IpXXS3S_1tfP?usp=share_link
Classification Dataset - The monitors present in the images can be grouped into 4
types, based on some screen characteristics. This is a separate dataset with the
monitors classified into those 4 separate types, a total of 250 images per class,
accounting for 1000 images.
Unlabelled Dataset - A total of 7000 unlabelled images. Can you use this data to
make your model better?