Fully On-Smartphone: Efficient Speed-Up of Sperm Tracking Algorithm for Point-of-Care Semen Analysis.

OBJECTIVE

To develop a lowcost, accurate, and fully-on-smartphone system for pointofcare semen analysis and to address the computational bottlenecks in dense sperm tracking on resourceconstrained devices.

METHODS

A portable optical attachment enabling microscopic video acquisition through consumer-grade smartphone cameras was developed. Initial deployment of the Joint Probabilistic Data Association Filter (JPDAF) on legacy mobile hardware revealed critical computational bottlenecks, where high-density semen samples ($\approx$ 80 million sperm/mL) induced exponential processing delays exceeding 3 hours due to exhaustive enumeration of feasible joint events in highdensity local regions. To overcome this limitation, the Global-Local Integrated JPDAF (GLIJPDAF) was proposed, featuring a twostage clustering strategy: (1) global clustering via a nondiagonal matrix partitioning algorithm to split the validation matrix into sparse submatrices isolating homogeneous association regions; (2) local clustering with a dynamic kmeans++ algorithm (adaptive k based on local non-zero density) to cap enumeration complexity.

RESULTS

Validation across four smartphone platforms using 90 patient samples (50-250 million sperm/mL) demonstrated that GLIJPDAF achieved a mean concentration error of 0.84 million/mL and a mean motility error of 0.74%. Processing times per 3s video remained under 120s on all devices.

CONCLUSION

Integrating GLIJPDAF into a smartphone platform enables rapid, accurate pointofcare semen analysis in highdensity samples on resourcelimited devices.

SIGNIFICANCE: This accessible pointofcare solution has the potential to broaden male infertility screening in lowresource settings.