TY - GEN
T1 - 'Digital Washing' of Semen Time-Lapse Images
AU - Alkhoury, Ludvik
AU - Sivri, Atilla
AU - Choi, Ji Won
AU - Bopp, Justin
AU - Anouna, Albert
AU - Henkel, Andreas W.
AU - Vermilyea, Matthew
AU - Kam, Moshe
N1 - Publisher Copyright:
© 2025 IEEE.
PY - 2025
Y1 - 2025
N2 - We introduce a supervised learning method for sperm cell detection in time-lapse images collected from raw human semen samples. This method uses a set of geometric features extracted from moving sperm cells (class SM) to identify immotile sperm cells (class SD) which typically share the same geometric features of moving sperm cells. It thus separates immotile sperm cells from other types of non-sperm cells (class O) and debris. We refer to this selective identification and separation process as 'Digital Washing'. It was tested on images collected from fourteen male volunteers. We compare the performance of the proposed method to that of two alternative methods, namely, the detection method of Urbano and his co-workers (2017) and YOLOv5 VISEM-Tracking (2023). Comparison criteria included precision, recall, and Fβ-scores. The proposed method provided precision of 0.82 ± 0.15, recall of 0.92 ± 0.03, F0.5-score of 0.83 ± 0.13, F1-score of 0.86 ± 0.09, and F2-score of 0.89 ± 0.05.
AB - We introduce a supervised learning method for sperm cell detection in time-lapse images collected from raw human semen samples. This method uses a set of geometric features extracted from moving sperm cells (class SM) to identify immotile sperm cells (class SD) which typically share the same geometric features of moving sperm cells. It thus separates immotile sperm cells from other types of non-sperm cells (class O) and debris. We refer to this selective identification and separation process as 'Digital Washing'. It was tested on images collected from fourteen male volunteers. We compare the performance of the proposed method to that of two alternative methods, namely, the detection method of Urbano and his co-workers (2017) and YOLOv5 VISEM-Tracking (2023). Comparison criteria included precision, recall, and Fβ-scores. The proposed method provided precision of 0.82 ± 0.15, recall of 0.92 ± 0.03, F0.5-score of 0.83 ± 0.13, F1-score of 0.86 ± 0.09, and F2-score of 0.89 ± 0.05.
KW - Computer-aided semen analysis
KW - Digital Washing
KW - Human sperm imaging
KW - Machine learning in biomedical imaging
KW - Motile and immotile cells
KW - Sperm cell detection
KW - Sperm cell geometric features
KW - Supervised learning
UR - https://www.scopus.com/pages/publications/105022088738
UR - https://www.scopus.com/pages/publications/105022088738#tab=citedBy
U2 - 10.1109/MLSP62443.2025.11204344
DO - 10.1109/MLSP62443.2025.11204344
M3 - Conference contribution
AN - SCOPUS:105022088738
T3 - IEEE International Workshop on Machine Learning for Signal Processing, MLSP
BT - 35th IEEE International Workshop on Machine Learning for Signal Processing
PB - IEEE Computer Society
T2 - 35th IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2025
Y2 - 31 August 2025 through 3 September 2025
ER -