How many labels do I need? Self-supervised learning strategies for multiple blood parasites classification in microscopy images
Mancebo-Martin, R.; Lin, L.; Dacal, E.; Luengo-Oroz, M.; Bermejo-Pelaez, D.
Show abstract
AO_SCPLOWBSTRACTC_SCPLOWBloodborne parasitic diseases such as malaria, filariasis or chagas pose significant challenges in clinical diagnosis, with microscopy as the primary tool for diagnosis. However, limitations such as time-consuming processes and the dependence on trained microscopists is critical, particularly in resource-constrained settings. Deep learning techniques have shown value to interpret microscopy images using large annotated databases for training. In this work, we propose a methodology leveraging self-supervised learning as a foundational model for blood parasite classification. Using a large unannotated database of blood microscopy images, the model is able to learn important image representations that are subsequently transferred to perform parasite classification of 11 different species of parasites requiring a smaller amount of labeled data. Our results show enhanced performance over fully supervised approaches, with [~]100 labels per class sufficient to attain an F1 score of [~]0.8. This approach is promising for advancing in-vitro diagnostic systems in primary healthcare settings.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Single-Cell Multi-Modal GAN (scMMGAN) reveals spatial patterns in single-cell data from triple negative breast cancer 90%
- Mistic: an open-source multiplexed image t-SNE viewer 90%
- A machine learning aided global diagnostic and comparative tool to assess effect of quarantine control in Covid-19 spread 89%
Similar papers in this journal
- Deep Learning Classification of Lipid Droplets in Quantitative Phase Images 95%
- Detection and measurement of butterfly eyespot and spot patterns using convolutional neural networks 94%
- Semantic Segmentation of HeLa Cells: An Objective Comparison between one Traditional Algorithm and Three Deep-Learning Architectures 94%
"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.