"The AMIL model showed the best performance at 5x magnification with an average AUC of 0.605, but results indicate limitations in detecting TP53 mutations."
"At 10x magnification, the AMIL model improved to an average AUC of 0.711, while the AdMIL model showed a reasonable but lower score of 0.624."
"The attention mechanism of the AMIL model at 10x focused more on inner patches, suggesting morphological patterns may be more relevant at this magnification level."
"The original AdMIL model failed to highlight relevant regions at 10x, pointing to isolated patches instead, indicating limitations in its detection capabilities."
At 5x magnification, the AMIL model achieved an AUC of 0.605, indicating poor performance in detecting TP53 mutations. This level does not provide discernible evidence of such mutations. Conversely, at 10x magnification, the AMIL model's performance improved, reaching an AUC of 0.711, while the AdMIL model scored 0.624. The AMIL model demonstrated effective patch attention mechanisms, focusing on inner patches, unlike the AdMIL model, which struggled to identify relevant excitation patches, instead highlighting isolated areas. Overall, magnification levels significantly influenced mutation detection capabilities.
Read at Hackernoon
Unable to calculate read time
Collection
[
|
...
]