Fig. 2

Study profile on the new smartphone app trained with original sonographic gallbladder images and smartphone photos taken by seven radiologists. Seven radiologists used smartphones to take pictures of the original sonographic gallbladder images. Smartphone photos were preprocessed before being inputted into the model for training. Through the fivefold cross-validation method, the new model was trained and finally outputted the predictions and probabilities for the test images. The optimized smartphone app outputted both the heatmap and the prediction probability at the same time