AI-assisted Diagnosis for a Pathology Lab
Faster diagnosis of the disease from histopathology images.
Achieved precision of 82% to 95% by reducing the analysis time from 15 minutes to ~4 minutes.
SITUATION
A Pathology lab wanted a Decision Support System to assist in the diagnosis of disease from histopathology images.
ADDITIONAL REQUIREMENTS
The developed DSS had to focus on:
- Data labeling: Label the data in coordination with the pathologist.
- Infection identification and classification: Analyze the WSIs scanned at 20x/40x resolution using a DNN model to identify and classify the infection.
- Application development: Development of an easy-to-use application for pathologists and their assistants.
CHALLENGES
- The DSS in the market failed to achieve high precision, resulting in multiple errors.
- The use of different scanners, stain coloring, and reactivity led to color variation in the histopathology images.
HIGHLIGHTS
- Labeled the data with the help of a microbiology professional under the guidance of the client pathologist.
- Extensive experimenting and training of convoluted neural networks (CNNs) from scratch (on TensorFlow) for multi-class classification using the labeled dataset on high-end GPU machines.
- On-premise deployment to ensure HIPAA compliance.
BENEFITS:
- Achieved precision of 82% to 95% on 4 classes.
- Reduced the analysis time from 15 minutes to ~4 minutes, enabling the lab to scale up by 400% of the existing caseload.