Deep Learning-Based Blood Abnormalities Detection as a Tool for VEXAS Syndrome Screening

Area of Science:

• Hematology
• Computational Biology
• Genetics

Background:

• VEXAS syndrome, a recently described condition (2020), is caused by UBA1 gene mutations and presents with diverse clinical and hematological features.
• Current diagnostic criteria for VEXAS lack specificity, making early differentiation from other inflammatory conditions challenging.
• Identifying unique dysplastic features in peripheral blood polymorphonuclear cells (PMNs) is crucial for improved VEXAS screening.

Purpose of the Study:

• To identify specific dysplastic features in peripheral blood polymorphonuclear cells (PMNs) indicative of VEXAS syndrome.
• To develop and validate a deep learning algorithm for the automated detection of these VEXAS-associated PMN abnormalities.
• To assess the utility of computer-assisted analysis in guiding molecular testing for VEXAS.

Main Methods:

• A multicentric dataset of 9514 annotated PMN images was compiled, including VEXAS patients (UBA1 mutated) and controls (UBA1 wildtype myelodysplastic and cytopenic).
• Statistical analysis was performed to identify significant PMN abnormalities differentiating VEXAS from controls.
• A convolutional neural network (CNN) was employed for automated, multilabel classification of PMN features.

Main Results:

• Significant differences in pseudo-Pelger forms, nuclear spikes, vacuoles, and hypogranularity were observed in PMNs from VEXAS patients compared to controls.
• The automated detection system achieved high performance with AUCs ranging from 0.85 to 0.97 and a F1-score of 0.70.
• A VEXAS screening score, derived from the model, demonstrated 0.82 sensitivity and 0.71 specificity in predicting UBA1 mutational status.

Conclusions:

• Computer-assisted analysis of peripheral blood smears can significantly aid in identifying VEXAS syndrome suspects.
• The developed deep learning approach assists hematologists in directing diagnostic suspicions and prioritizing molecular testing.
• This technology enhances the efficiency of VEXAS diagnosis by enabling targeted investigations.