![]() ![]() In the CNS, reactive astrocytosis can acquire morphological features resembling malignant giant cells, even in inflammatory conditions like multiple sclerosis 21. This is in contrast to, for example, solid epithelial tumors, where de-differentiated malignant cells may share some morphological features with the tissue of origin (e.g., gland-like structures), but pre-existing surrounding tissues do not significantly change their characteristics towards a malignant appearance. Tumor cells can resemble their normal counterparts (astrocyte-like differentiation) and non-malignant astrocytes may assume phenotypes resembling malignant cells (reactive astrocytosis) 20. However, it is difficult to distinguish reactive astrocytes (cells from central nervous system) from almost identically-looking tumor (glioma) cells in this use case. This typically requires specific strategies to achieve satisfying results including “educated” crowds with prior knowledge (e.g., medical students) and/or an advanced work setting, such as direct contact between requester and crowdworker 16, 17, 18.Īccurate cell detection is critical to analyze brain tumor microenvironments 19. A third type of task complexity concerns required skills for image annotation, such as delineation of anatomical or pathologically altered structures. Secondly, complexity strongly increases with higher number of classes e.g., school pupils annotated cells in a competitive format with levels from “mild” (annotating a single cell type) step-wise adding further cell types up to “supercharger” level, with decreasing accuracy in higher levels 15. First, semantic content and visual appearance have a strong influence on how crowdworkers perceive task complexity 14. Expanding crowdsourcing towards more complex problems requires several considerations. In microscopic images, crowds were asked for scoring of unambiguously stained cells or annotation of cell nuclei 11, 12, 13, representing tasks of low complexity. In the case of paid contributions, compensation schemes can influence performance 7 and thus may have impact on quality of results.Ĭrowdsourcing has been widely used in cancer research 8, 9, 10. It is recommendable to test task design in pilot studies to prove its reliability. Further, tasks should allow timely completion, preferably within less than 10 min 6. Crowd instructions should use simple English without complex scientific terms and provide illustrated or animated explanations in the qualification phase. Therefore, strategies for successful task design include a focus on simple tasks and qualification phases with performance tests 5. ![]() In this constellation, neither the contributors’ educational backgrounds, nor their environmental conditions are known, as there is no direct contact to the crowd. It considers the collaborative solution of problems by several workers with heterogeneous domain knowledge, in the format of participatory online activity 4. Similar content being viewed by othersĬrowdsourcing addresses the urgent need for training data in machine learning (ML), as it has been shown that crowdlabels can be feasible to train convolutional neural networks (CNNs) 1, 2, 3. Our work confirms feasibility of crowdsourcing to generate labels suitable for training of machine learning tools in the challenging and clinically relevant use case of glioma microenvironment. The networks resulted in acceptable \(F_1\) scores up to 0.69 for YOLOv8 on average and indicated first evidence for transferability to images lacking tumor markers, especially in IDH-wildtype glioblastoma. A crowd of 712 workers defined aggregated point annotations in 235 images with an average \(F_1\) score of 0.627 for majority vote. On the base of a YOLO convolutional neural network architecture, we used these consensus labels for training with different image representations regarding colors, intensities, and immmunohistochemical marker combinations. We applied majority or weighted vote and validated them against ground truth in the final setting. In four pilot experiments, we iteratively developed a task design enabling high-quality annotations by crowdworkers on Amazon Mechanical Turk. ![]() This requires multiplexed immunofluorescence microscopy due to diffuse invasiveness and exceptional similarity between glioma cells and reactive astrocytes. Herein, we broaden its scope to the previously unsolved challenging task of glioma cell detection. Crowdsourcing has been used in computational pathology to generate cell and cell nuclei annotations for machine learning. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |