Fig 2 - 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
... follow similar patterns as those of YOLOv11. In comparing optimizers, YOLOv8 demonstrated greater stability when using SGD rather than AdamW. For batch size, we experimented with 8, 16, 32, and 64, finding that smaller batch sizes yielded better performance. Consequently, a batch size of 8 was chosen for the final model. The confusion matrix in Fig. 22 illustrates how the number of misclassified normal cells surpasses that of the small and nano versions of YOLOv11; otherwise, the model is performing considerably well and ...
Context 2
... this section, we evaluate ResNet50's behavior and performance metrics. Using fine-tuning, the model's training accuracy settled at a peak of 99.2%. Meanwhile, the validation accuracy grew to 99%, and the test dataset evaluation achieved 99%. The visualization of the training and validation accuracy curves in Fig. 23 highlights the learning enhancement and growth along the epochs. There were some variations at the beginning of the validation curve that were reduced as the training process progressed, and there was almost no gap between the training and validation by the hundredth epoch, which also applies to the training and validation losses in ...
Context 3
... in Fig. 23 highlights the learning enhancement and growth along the epochs. There were some variations at the beginning of the validation curve that were reduced as the training process progressed, and there was almost no gap between the training and validation by the hundredth epoch, which also applies to the training and validation losses in Fig. 24. On a different note, the confusion matrix of ResNet50 in Fig. 25 illustrates that there is a percentage of healthy white blood samples that was misidentified as cancerous. However, it identifies all of the blast cells correctly. ...
Context 4
... the epochs. There were some variations at the beginning of the validation curve that were reduced as the training process progressed, and there was almost no gap between the training and validation by the hundredth epoch, which also applies to the training and validation losses in Fig. 24. On a different note, the confusion matrix of ResNet50 in Fig. 25 illustrates that there is a percentage of healthy white blood samples that was misidentified as cancerous. However, it identifies all of the blast cells correctly. ...
Context 5
... training and validation losses exhibited a similar trend but in the opposite direction, declining over time with some sharp variations for the validation at the start of the process. It is worth noting that, compared to ResNet50, Inception-ResNet-v2 demonstrated a more stable training process. The confusion matrix of the model is visualized in Fig. 28, which highlights the model's strong capability in distinguishing between cancerous and normal samples with very few misclassifications. Table IV highlights the key findings of this study by evaluating the algorithms on the test dataset using various evaluation metrics. To sum up, it is notable that InceptionResNet-v2 comes first in ...
