Medical records processing · Production

Advent Health Partners uses active learning and model-assisted labeling to speed up medical record processing with Labelbox

The problem

Extracting entities from millions of multi-format medical records (PDFs, faxes, scans) from diverse providers with no consistent formatting was time-intensive and expensive, creating a bottleneck for training clinical ML models.

Workflow diagram · grounded in source
1
Medical records ingested
trigger
“these records typically come from numerous different providers, hospitals, insurance companies, with no consistent formatting and contain image and text files of varying formats such as PDFs, faxes, scans, etc.”
2
OCR and NLP entity extraction
ai_action
“leveraging the latest advances in optical character recognition (OCR) on these medical records and applying AI through the use of natural language processing (NLP) in the form of entity extraction”
3
Page classification model
ai_action
“they took paper records and automatically tagged emergency department records, discharge notes, and types of documents that human reviewers can check within the Labelbox interface and review for accuracy”
4
Human accuracy review
human_review
“Their in-house team of domain experts would review both confident model predictions to prevent errors and utilize uncertainty sampling”
5
Entropy-based active learning
ai_action
“they evaluated their existing model over unlabeled samples and calculated the entropy of the output classification vector. The AHP team grouped these calculations into two buckets: one with low entropy (meaning where the model was confid…”
6
Semi-supervised class balancing
feedback_loop
“they used the earlier version of their model to create a semi-supervised model that would balance the classes within their unlabeled data which resulted in better model performance”
7
Labeled data delivered to ML models
output
“The AHP data science team's primary goal is to analyze and feed this vast amount of information into their clinical machine learning models for focused review, and to examine whether or not the documentation has all the necessary element…”
Reported outcome

By adopting Labelbox's active learning and model-assisted labeling, AHP reduced the average time per label from 13 seconds to 8 seconds and cut 25 hours off each specific labeling task, while also achieving better model performance through semi-supervised class balancing.

Reported metrics
Time per label (before)13 seconds
Time per label (after)8 seconds
Hours saved per labeling task25 hours
Labeling process speedupdramatically speed up the process
Reported stack
LabelboxOCRNLP
Source
https://labelbox.com/customers/advent-health-partners-customer-story
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

By adopting Labelbox's active learning and model-assisted labeling, AHP reduced the average time per label from 13 seconds to 8 seconds and cut 25 hours off each specific labeling task, while also achieving better mod…

What tools did this team use?

Labelbox, OCR, NLP.

What results were reported?

Time per label (before): 13 seconds; Time per label (after): 8 seconds; Hours saved per labeling task: 25 hours; Labeling process speedup: dramatically speed up the process (source-reported, not independently verified).

How is this medical records processing AI workflow structured?

Medical records ingested → OCR and NLP entity extraction → Page classification model → Human accuracy review → Entropy-based active learning → Semi-supervised class balancing → Labeled data delivered to ML models.

Related medical records processing cases