Use this link to cite:
http://hdl.handle.net/2183/39440 Detección y clasificación automática de larvas de erizo de mar en imágenes microscópicas mediante aprendizaje profundo
Loading...
Identifiers
Publication date
Authors
Millares Goy, Manuel
Other responsabilities
Universidade da Coruña. Facultade de Informática
Journal Title
Bibliographic citation
Type of academic work
Academic degree
Abstract
[Resumen]: Los erizos de mar, particularmente en su fase de larva pluteus, son fundamentales para investigaciones ambientales y toxicológicas debido a su morfología característica y translúcida. Estas características los hacen indicadores ideales para estudiar el impacto de contaminantes en el medio marino. Tradicionalmente, la evaluación de su desarrollo y respuesta a toxinas se ha realizado manualmente, un proceso que, aunque detallado, resulta lento y propenso a variabilidad. Este trabajo introduce métodos de aprendizaje profundo para la detección y clasificación automática de larvas de erizos de mar en imágenes microscópicas, ofreciendo dos enfoques novedosos. El primer enfoque es un método secuencial que realiza primero la segmentación y luego la clasificación de las larvas. El segundo enfoque utiliza un modelo integrado que efectúa simultáneamente la segmentación y clasificación, optimizando el proceso y reduciendo los errores inherentes al método secuencial. Los resultados obtenidos demuestran que ambos métodos superan significativamente la precisión y eficiencia de las técnicas manuales, con el modelo integrado mostrando particularmente un rendimiento superior. Esta mejora en la precisión y la reducción del tiempo de análisis son cruciales frente al creciente número de contaminantes en los océanos y su impacto
potencialmente nocivo en la biota marina. La implementación de estas tecnologías no solo optimiza los estudios ecotoxicológicos sino que también promueve una monitorización ambiental más efectiva y eficiente, proveyendo herramientas necesarias para la rápida detección y respuesta frente a la contaminación marina. Este avance representa un paso significativo hacia la automatización en la ciencia marina, ofreciendo una metodología confiable y reproducible para el estudio de la embriogénesis y la toxicidad en erizos de mar.
[Abstract]: Sea urchins, particularly in their pluteus larval stage, are critical for environmental and toxicological research due to their distinctive and translucent morphology. These characteristics make them ideal indicators for studying the impact of pollutants in the marine environment. Traditionally, the assessment of their development and response to toxins has been conducted manually, a process that, although detailed, is slow and prone to variability. This work introduces deep learning methods for the automatic detection and classification of sea urchin larvae in microscopic images, offering two novel approaches. The first approach is a sequential method that performs segmentation and then classification of the larvae. The second approach uses an integrated model that simultaneously performs segmentation and classification, optimizing the process and reducing the errors inherent in the sequential method. The results obtained demonstrate that both methods significantly surpass the accuracy and efficiency of manual techniques, with the integrated model particularly showing superior performance. This improvement in accuracy and reduction in analysis time are crucial in the face of the increasing number of pollutants in the oceans and their potentially harmful impact on marine biota. The implementation of these technologies not only optimizes ecotoxicological studies but also promotes more effective and efficient environmental monitoring, providing the necessary tools for rapid detection and response to marine pollution. This advancement represents a significant step towards automation in marine science, offering a reliable and reproducible methodology for studying embryogenesis and toxicity in sea urchins.
[Abstract]: Sea urchins, particularly in their pluteus larval stage, are critical for environmental and toxicological research due to their distinctive and translucent morphology. These characteristics make them ideal indicators for studying the impact of pollutants in the marine environment. Traditionally, the assessment of their development and response to toxins has been conducted manually, a process that, although detailed, is slow and prone to variability. This work introduces deep learning methods for the automatic detection and classification of sea urchin larvae in microscopic images, offering two novel approaches. The first approach is a sequential method that performs segmentation and then classification of the larvae. The second approach uses an integrated model that simultaneously performs segmentation and classification, optimizing the process and reducing the errors inherent in the sequential method. The results obtained demonstrate that both methods significantly surpass the accuracy and efficiency of manual techniques, with the integrated model particularly showing superior performance. This improvement in accuracy and reduction in analysis time are crucial in the face of the increasing number of pollutants in the oceans and their potentially harmful impact on marine biota. The implementation of these technologies not only optimizes ecotoxicological studies but also promotes more effective and efficient environmental monitoring, providing the necessary tools for rapid detection and response to marine pollution. This advancement represents a significant step towards automation in marine science, offering a reliable and reproducible methodology for studying embryogenesis and toxicity in sea urchins.
Description
Keywords
Inteligencia artificial Aprendizaje profundo Erizo de mar Paracentrotus lividus Imagen microscópica Visión por computador Segmentación de imágenes Clasificación de marcadores Clasificación de instancias Artificial intelligence Deep learning Sea urchin Paracentrotus lividus Microscopic image Computer vision Image segmentation Image classification Instance classification
Editor version
Rights
Todos os dereitosreservados. Neste caso prohíbese a reprodución, transformación,
distribución e comunicación pública da obra por terceiros. En cambio, permítese a
súa visualización e a descarga dunha copia privada para o uso persoal.







