BlotDx: A deep learning tool for Western blot-based diagnostics

Area of Science:

• Medical Diagnostics
• Artificial Intelligence in Medicine
• Bioinformatics

Background:

• Western blot (WB) is the standard for herpes simplex virus (HSV-1/HSV-2) serology.
• Manual interpretation of WB is time-consuming and prone to variability.

Purpose of the Study:

• To introduce BlotDx, a deep learning tool designed to assist in HSV WB interpretation.
• To enhance the efficiency and diagnostic consistency of HSV serology.

Main Methods:

• BlotDx employs a two-stage deep learning approach: instance segmentation/object detection for blot identification and classification models for serostatus determination.
• Three distinct classifiers were developed, varying in information flow between stages.
• The study utilized a primary dataset of 926 blot pairs and a validation dataset of 185 blot pairs.

Main Results:

• BlotDx achieved high diagnostic accuracy in both datasets, validated against expert human review.
• In the primary dataset, accuracy was 98.8% for HSV-1 and 98.9% for HSV-2.
• In the validation dataset, accuracy was 97.3% for HSV-1 and 96.2% for HSV-2.

Conclusions:

• AI tools like BlotDx can serve as valuable diagnostic assistants for image-based assays.
• The two-stage deep learning approach is scalable and can transform traditional diagnostic workflows.
• BlotDx has the potential to reduce costs and increase efficiency in HSV serology and other diagnostic applications.