Access and Collaborate on Structured and Unstructured Clinical Data
Seamlessly integrate medical imaging data with patient’s phenotypic clinical data to unlock holistic patient insights that help drive better intervention and care coordination.
UNLOCK THE POWER
OF UNSTRUCTURED
MEDICAL IMAGING DATA
WITH SNOWFLAKE SNOWPARK
Solution Overview
Overview
Digital Imaging and Communications in Medicine (DICOM) is the international standard for medical imaging information, like x-rays or a CT scan, along with medical record metadata. To generate holistic patient insights, parse medical DICOM images on Snowflake using Snowpark to create a DICOM metadata repository that includes clinical data. Then train and deploy machine learning models using Snowpark to classify medical images based on diagnosis for a health condition, such as pneumonia on a chest x-ray.
Opportunity
According to the NIH, about 80% of medical data remains untapped and unstructured, such as medical images, which makes it difficult to connect with medical research and artificial intelligence for healthcare. For example, chest x-ray imaging is the most frequently used method for diagnosing pneumonia—yet examining x-rays can be challenging due to subjective variability, time needed to analyze images and a lack of trained professionals.
Solution
AI is a potential solution to improve detection of health conditions like pneumonia. By parsing DICOM image data and training a ML model on Snowflake, healthcare professionals can diagnose health conditions faster and more accurately.
Solution Architecture
DICOM Images are stored in cloud storage buckets through integration with PACS or Vendor Neutral Archive System (VNA), and are accessed through external stage integration with Snowflake. Through Snowpark’s unstructured file access capability, pydicom—an open source python package—accesses DICOM files and extracts metadata information from images before storing them as a DICOM repository on Snowflake. When combined with clinical data on Snowflake, this provides a holistic view of the patient's health journey.
Snowpark python packages read and convert DICOM images into vectorized data. A Python Tensorflow model is built to read the vectorized medical image data and predict the possibility of pneumonia in chest x-rays. The model is then exported and deployed for inference in the form Snowflake User Defined Function (UDF). This UDF can then be called through a stored procedure for batch inference or through external API calls.
Benefits
