Fig 1 - available via license: Creative Commons Zero 1.0
Content may be subject to copyright.
Source publication
Thousands of individuals succumb annually to leukemia alone. As artificial intelligence-driven technologies continue to evolve and advance, the question of their applicability and reliability remains unresolved. This study aims to utilize image processing and deep learning methodologies to achieve state-of-the-art results for the detection of Acute...
Contexts in source publication
Context 1
... implementation is divided into several phases, as illustrated in Fig. 1. The first phase involved data preparation, where image segmentation techniques were applied to isolate the relevant elements. Additionally, image rescaling was done to ensure that each model receives the right shape for the input layer while data augmentation was utilized to assist with the data variety and enhancing the convergence. ...
Context 2
... also conducted experiments with different optimizers and hyperparameters. Figures 9, 10, and 11 exhibit the model's performance when leveraging AdamW optimizer and the default 0.000714 YOLO learning rate during 100 epochs on batches of size 16. AdamW optimizer shows more unsteady performance than SGD in the accuracy and validation loss graphs. ...
Context 3
... exhibited a similar but enhanced behavior compared to YOLOv11n. The optimal performance on YOLOv11n was reached with SGD optimizer, 32 batch size, and 0.001 learning rate, and the model was trained for 50 epochs. Fig. 12 clarifies how the accuracy value followed the same trend as YOLOv11n, where it rose with some random variations at the beginning indicating that the model was still learning the new data patterns. The graph became more stable by the end of the training process, achieving 98.6% validation accuracy. In addition, the test accuracy reached ...
Context 4
... addition, the test accuracy reached 98.2%, surpassing that of Yolov11n by 0.9. Figures 13 and 14 demonstrate the There was a slight improvement in the confusion matrix as well, which can be illustrated in Fig. 15, such that the rate of images of healthy white blood cells misclassified as cancer was less than that of YOLOv11n since it fell from 0.10 to 0.07. ...
Context 5
... still learning the new data patterns. The graph became more stable by the end of the training process, achieving 98.6% validation accuracy. In addition, the test accuracy reached 98.2%, surpassing that of Yolov11n by 0.9. Figures 13 and 14 demonstrate the There was a slight improvement in the confusion matrix as well, which can be illustrated in Fig. 15, such that the rate of images of healthy white blood cells misclassified as cancer was less than that of YOLOv11n since it fell from 0.10 to ...
Context 6
... the other hand, Figures 12, 13, and 14 visualize the model's performance when trained using the AdamW optimizer. Although it attained a higher validation accuracy of 98.8%, the training process showed considerable fluctuations, reflecting instability and a possible risk of overfitting. ...
