![]() Data-level methods include oversampling through various data augmentation techniques to enlarge the minority class image set.ĭata augmentation is usually achieved using two types of approaches. Parameter-level methods alter the learning or decision process by assigning higher weights and often undergo a precarious process of weight determination. In existing studies 7, 8, approaches to handle data imbalance can be categorised into parameter-level and data-level. Therefore, in order to achieve a model that is applicable in real world settings, it is important for the model to be trained on all the classes equitably. Furthermore, minority classes often constitute more than half the number of total classes in a biomedical image dataset. The underrepresentation of minority classes causes model training to be heavily biased towards the majority classes, which in turn leads to a significant performance drop of the trained model on minority classes. Consequently, samples from majority classes are frequently observed and those from minority classes rarely encountered. ![]() In data imbalance settings, some classes have far higher numbers of samples compared to other classes in the dataset. Since data collection in biomedical studies is restricted by the nature of the biological phenomena, data imbalance is a common issue in CNN based classification 7. However, the efficacy of CNN based classifiers is logarithmically proportional to the amount of training data 7. Likewise, in non-nucleated blood cells (erythrocytes/Red Blood Cells (RBCs)), morphological variations due to storage need to be identified for the prediction of blood quality for life-saving blood transfusions 2.Ĭonvolutional Neural Networks (CNNs) have exhibited state-of-the-art performances to identify cellular morphologies 2, 3, 4, subcellular localisations 5 and cell-cycle phases 1, 6. Classification of cell-cycle phases in nucleated blood cells (lymphocytes) is vital for diagnostic and prognostic research studies of pathological conditions and impacts clinical decision making 1. ![]() Classification performance evaluated using F1-score shows that our proposed approach outperforms existing methods on the same datasets.īlood cells undergo various morphological changes as they progress in their life cycle or undergo the impact of environmental factors. The method was evaluated on two publicly available datasets of immortalised human T-lymphocyte cells and Red Blood Cells. We also present a minority class focussed sampling strategy, which allows effective representation of minority class samples produced by all three data augmentation techniques and contributes to the classification performance. ![]() This study presents an imbalanced blood cells classification method that utilises Wasserstein divergence GAN, mixup and novel nonlinear mixup for data augmentation to achieve oversampling of the minority classes. However, real-world data often suffers from the data imbalance problem, owing to which the trained classifier is biased towards the majority classes and does not perform well on the minority classes. Deep learning and more specifically Convolutional Neural Networks have achieved state-of-the-art performance on various biomedical image classification problems. Automated classification systems help avoid subjective outcomes and are more efficient. Due to progression in cell-cycle or duration of storage, classification of morphological changes in human blood cells is important for correct and effective clinical decisions.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |