Clasificador del estado de la membrana acrosomal en espermatozoides mediante redes neuronales convolucionales
Artificial intelligence (AI) has transformed the livestock industry, becoming a fundamental tool for analyzing sperm quality in cattle, contributing to the economic and productive development of the country. This project employs convolutional neural networks (CNNs) to develop a model that classifies...
Guardado en:
| Autores principales: | , |
|---|---|
| Otros Autores: | |
| Formato: | Tesis/Trabajo de grado - Monografía - Pregrado |
| Lenguaje: | Español |
| Publicado: |
Universidad Antonio Nariño
2025
|
| Materias: | |
| Acceso en línea: | https://repositorio.uan.edu.co/handle/123456789/12404 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Artificial intelligence (AI) has transformed the livestock industry, becoming a fundamental
tool for analyzing sperm quality in cattle, contributing to the economic and productive development
of the country. This project employs convolutional neural networks (CNNs) to develop a model
that classifies the acrosomal membrane status in bull sperm, providing additional information that
supports the assessment of artificial insemination success in the field.
The Biotechnology Laboratory at the Antonio Nariño University, Popayán campus, has
advanced equipment for analyzing sperm quality in animals but lacks compatible software to better
visualize and understand the results. To address this need, a classifier based on YOLOv5, a pretrained CNN, was developed to identify the acrosomal status in bull sperm through image
processing and machine learning.
For the analysis, a double staining of samples with FITC-PNA and PI was used, with green
and red fluorochromes that allow morphological differentiation of the acrosome and sperm head.
Cells with both green and red fluorescence are classified as AM (Bad Acrosome), while those with
only red fluorescence are classified as AB (Good Acrosome). A database was created with 300
high-resolution images (1280 x 1024 pixels) captured with the Nikon Eclipse Ti2-U microscope
and Basler camera.
The YOLOv5 model was implemented in Visual Studio Code using Python, resizing the
images to 640 x 640 pixels and setting the batch size and the number of epochs to 8 and 80,
respectively. The AM and AB classes were defined in the model to make predictions on the
acrosomal status of the sperm.
The results showed that the model achieved 97% accuracy in the AM class, excelling in
recognizing this predominant class. For the AB class, accuracy was 79%, affected by factors such
IX
as a smaller data quantity and fluorescence loss. In conclusion, the classifier accurately identifies
sperm classes, providing a reliable index on the acrosomal status of sperm cells. |
|---|
