Submitter Withdrawn ESA-SRB-NZSE-CaSR 2026 in conjunction with ENSA

Preliminary experience on the use of artificial intelligence-assisted sperm search in surgical sperm retrieval. (#472)

Saud I Sharif Hussain 1 , Khalid Al Muqati 1 , Jaffar Ali 1 2
  1. Dr Sulaiman ALHabib Hospital, Riyadh, NAJD, Saudi Arabia
  2. Faculty of Medicine, UCSI University, Port Dickson, Negri Sembilan, Malaysia

The objective was to evaluate the SpermSearchAITM software. Observation 1: The observation dish was identical to manufacturer's. Observation 2 2: The dish was flooded with a thin layer of medium containing a mixture of spermatozoa and testicular tissue cells. Observation 1 attempted to detect a clean specimen of extremely low concentrations of washed clean spermatozoa (cS) held in a HEPES-buffered medium devoid of any cells. Subsequent investigations repeated the experiment with TESA (TE) specimen. Observation 1 noted the AI detected spermatozoa faster, but the human eye detected significantly more (p=0.0085) spermatozoa, mean spermatozoa counted (MSCs):13.8 vs 8.4 respectively, n=20, correlation coefficient r=0.9285, p<0.0001, while the mean false positive observations (MFOs) were significantly higher in AI-group, mean 0.2 vs 0.0, p<0.05, with no MFOs in the human. In TE specimen, the MSCs were similar in human and AI observations: 7.9 vs 8.0, p=0.7054, respectively, n=20, r=0.9558, p<0.0001; FPOs were similar to the previous observation, mean 3.0 vs 0.0, p<0.0001. TE specimen without dilution, mean spermatozoa counts were 3.35 vs 1.2, p<0.0001, for human and AI respectively, n=20, r=0.6485, p=0.0020, the FPOs were similar to previous experiments, mean 2.3 vs 0.0, p<0.0001); so were results of  corresponding diluted TE specimen, the difference between mean spermatozoa counts were not significant, 0.78 vs 0.70, p=0.1619, n=23, r=0.9221, p<0.0001, the FPOs were similar to previous experiments mean 1.3 vs 0.0, p<0.0001. When the observations of all experiments, except the first were combined, the mean of spermatozoa counts were significantly higher (p=0.0003) in the human compared to AI respectively: 3.9 vs 3.2, n=63, r=0.9353, p<0.0001, while the FPOs were similar to previous experiments, mean 2.2 vs 0.0, p<0.0001. In conclusion, AI detects spermatozoa faster but with significantly high false positive observations than the human but the human counts significantly more spermatozoa with no false positive observations.