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

Western blot (WB) is the gold standard for herpes simplex virus (HSV-1/HSV-2) serology but requires manual interpretation by trained laboratory personnel, which is time-consuming and variable. This study presents BlotDx, a deep learning tool to assist in HSV WB interpretation to improve efficiency and diagnostic consistency.

BlotDx uses a two-stage approach: (1) instance segmentation or object detection to identify blots from input images, and (2) classification models to determine positive or negative serostatus for HSV-1 and HSV-2, excluding indeterminate results. We developed three classifiers that differ in how information flows between stages. The primary dataset consisted of 926 blot pairs derived from samples collected between 2016 and 2017 and photographed in 2018; and a second institutional validation dataset containing 185 blot pairs from samples collected between 2019 and 2024 and photographed in a different laboratory space in 2025.

Evaluated against the ground truth of expert human review by three independent technicians, BlotDx demonstrated high diagnostic accuracy by 5-fold cross validation in both the primary dataset (98.8% for HSV-1, 95% confidence interval (CI): 97.9%-99.3%; and 98.9% for HSV-2, 95% CI: 98.0%-99.4%), and in the institutional validation dataset (97.3% for HSV-1, 95% CI: 93.8%-98.8%; and 96.2% for HSV-2, 95% CI: 92.4%-98.1%).

This study highlights the utility of AI as a diagnostic assistant for image-based assays like HSV Western blots, with potential applications in other diseases. The proposed two-stage approach, combined with modern deep learning techniques is scalable and offers a step towards transforming traditional diagnostic workflows by reducing costs and increasing efficiency.