Hello seniors and expertise,
@ptrblck am working on Chest X-rays and I want to detect and classify different thoric diseases. I trained a model on a Large NIH dataset but the AUC is not good. For improving the model I am going to retrain the trained model with fewer classes because the old dataset and new dataset classes are not matched 100%.
I have only 9 classes that are matched with each other.
For this is I’m confused that how I can retrain the model?
Question:
Is it possible to train the model?
How much is it worth to my model in terms of performance?
I trained my model on PNG images whether the new dataset is .dicom images which having numpy arrays. So Does it reduce the performance?
Please help regarding this.
Thanks