Existing API approaches usually independently leverage detection or classification models to distinguish allergic pollens from Whole Slide Images (WSIs). However, palynologists tend to identify pollen grains in a progressive learning manner instead of the above one-stage straightforward way. They generally focus on two pivotal problems during pollen identification. (1) Localization: where are the pollen grains located? (2) Classification: which categories do these pollen grains belong to? To perfectly mimic the manual observation process of the palynologists, we propose a progressive method integrating pollen localization and classification to achieve allergic pollen identification from WSIs. Specifically, data preprocessing is first used to cut WSIs into specific patches and filter out blank background patches. Subsequently, we present the multi-scale detection model to locate coarse-grained pollen regions (targeting at “pollen localization problem”) and the multi-classifiers combination to determine the fine-grained category of allergic pollens (targeting at “pollen classification problem”). Extensive experimental results have demonstrated the feasibility and effectiveness of our proposed method.
Existing pollen identification methods heavily rely on the scale and quality of pollen images. However, there are many impurities in real-world SEM images that should be considered. This paper proposes a collaborative learning method to jointly improve the performance of pollen segmentation and classification in a weakly supervised manner. It first locates pollen regions from the raw images based on the detection model. To improve the classification performance, we segmented the pollen grains through a pre-trained U-Net using unsupervised pollen contour features. The segmented pollen regions were fed into a deep convolutional neural network to obtain the activation maps, which were used to further refine the segmentation masks. In this way, both segmentation and classification models can be collaboratively trained, supervised by just pollen contour features and class-specific information. Extensive experiments on real-world datasets were conducted, and the results prove that our method effectively avoids impurity interference and improves pollen identification accuracy (86.6%) under the limited supervision (around 1000 images with image-level labels).
This paper presents an eye image segmentation-based computer-aided system for automatic diagnosis of ocular myasthenia gravis (OMG), called OMGMed. It provides great potential to effectively liberate the diagnostic efficiency of expert doctors (the scarce resources) and reduces the cost of healthcare treatment for diagnosed patients, making it possible to disseminate high-quality myasthenia gravis healthcare to under-developed areas. The system is composed of data pre-processing, indicator calculation, and automatic OMG scoring. Building upon this framework, an empirical study on the eye segmentation algorithm is conducted. It further optimizes the algorithm from the perspectives of “network structure” and “loss function”, and experimentally verifies the effectiveness of the hybrid loss function. The results show that the combination of “nnUNet” network structure and “Cross-Entropy + Iou + Boundary” hybrid loss function can achieve the best segmentation performance, and its MIOU on the public and private myasthenia gravis datasets reaches 82.1% and 83.7%, respectively. The research has been used in expert centers. The pilot study demonstrates that our research on eye image segmentation for OMG diagnosis is very helpful in improving the healthcare quality of expert doctors. We believe that this work can serve as an important reference for the development of a similar auxiliary diagnosis system and contribute to the healthy development of proactive healthcare services.