An integrated AI-microfluidic platform reveals the broad persistence and developmental potential of rare sperm in non-obstructive azoospermia
Chen, H.; Chen, P.; Xiao, W.; Wang, L.; Song, M.; Liu, X.; Shen, R.; Guo, S.; Li, J.; Zhao, W.; Mo, M.; Huang, C.; Xu, S.; Sun, Q.; Zhong, H.; Ye, L.; Xi, Y.; Chen, C.; Xiong, F.; Zhang, H.; Wang, X.
Show abstract
Non-obstructive azoospermia (NOA) represents the most severe form of male infertility, severely limiting a patient's prospects for biological fatherhood when surgical retrieval fails. However, the true biological limits of NOA remain obscured by the inherent limitations of conventional gamete recovery protocols: standard centrifugation frequently causes substantial cell loss, masking extremely rare sperm, while surgical interventions are constrained by spatial sampling biases. Here we report SpermSeek, an integrated AI-guided microfluidic platform for real-time, non-destructive isolation of single sperm directly from semen. Operating at scalable throughput (0.36 mL/h), the system achieves 98.3% detection precision and a 95.5% target encapsulation efficiency, suppressing background debris. In a 59-patient NOA cohort, SpermSeek detected morphologically identifiable sperm in 64.4% (38/59) of cases, spanning diverse genetic etiologies, including AZFb/c microdeletions, and severe histopathological phenotypes, such as Sertoli-cell-only syndrome (SCOS). Notably, among a sub-cohort of 41 patients who remained consistently sperm-negative despite prior medical or micro-TESE interventions, our platform identified gametes in 53.7% (22/41) of these cases. Comprehensive safety profiling in healthy human donors and wild-type mice confirmed that processed sperm retain high DNA integrity and epigenomic concordance (r=0.98), supporting transgenerational developmental stability in mice. Furthermore, in a 26-patient validation cohort, SpermSeek recovered rare sperm in 11 cases. Utilizing gametes from a subset (n=5), we demonstrated their capacity to support early human embryogenesis, yielding high-quality cleavage-stage embryos with confirmed genomic euploidy. This work establishes a highly sensitive framework for re-examining the biological limits of human spermatogenesis, laying the foundation to expand autologous reproductive options for patients refractory to conventional retrieval protocols.
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Optimized double emulsion flow cytometry with high-throughput single droplet isolation 94%
- Deterministic Cell Pairing with Simultaneous Microfluidic Merging and Sorting of Droplets 94%
- Dual Dean entrainment with volume ratio modulation for efficient co-encapsulation: Extreme single-cell indexing 94%
Similar papers in this journal
- At Home Detection of Ovarian Health Biomarker in Menstruation Blood 92%
- Spatiotemporal control of a multilayered co-axial flow in a 3D printed microchannel with cascaded nozzles 91%
- Non-contact microfluidic analysis of the stiffness of single large extracellular vesicles from IDH1-mutated glioblastoma cells 91%
Similar papers in this journal
- Deep Learning-Enabled Point-Of-Care Sensing Using Multiplexed Paper-Based Sensors 90%
- High-Sensitivity Pan-Cancer AI Assessment of Lymph Node Metastasis via Uncertainty Quantification 88%
- Improving Pre-eclampsia Risk Prediction by Modeling Individualized Pregnancy Trajectories Derived from Routinely Collected Electronic Medical Record Data 88%
Similar papers in this journal
- A Cascaded Droplet Microfluidic Platform Enables High-throughput Single Cell Antibiotic Susceptibility Testing at Scale 95%
- In vitro neutrophil functional assay in microliter whole blood for days-long extraction of donor-specific information 94%
- Microfluidic-enabled production of DNA barcoded APC library (MEDAL) for high throughput T cell epitope screening 92%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.