Publications by authors named "Rifat Edizkan"
Phys Eng Sci Med
September 2024
Article Synopsis
- Deep learning methods have shown great success in biomedical image processing, particularly for automatic diagnosis, but often prioritize accuracy over interpretability and appropriate data separation.
- Many studies shuffle data randomly for training and testing, which can lead to misleading accuracy and irrelevant feature learning since images from the same patient may be split across different sets.
- In contrast, models trained with strict patient-level separation demonstrate better performance on new patient images and provide clearer visualizations of relevant features, indicating a more reliable approach for improving real-life applicability.
