The experiment was run four times to evaluate the stability of training. The convergence of loss value in training was achieved to evaluate the training process. The preprocessed image was input into the well-trained networks, the probability was calculated, and a decision was output. The average and standard deviation (SD) of accuracy, precision, sensitivity, specificity and F1 score of four runs on test set were calculated, the definitions of which were given by:

F1 score is an index to measure the accuracy of binary classification model, which conveys the balance between precision and sensitivity. The effectiveness was also evaluated by drawing the receiver operating characteristic (ROC) curve and calculating the area under curve (AUC).

In addition, it is difficult to tell how neural network works, so the occlusion testing, a kind of visualization method, was applied to reveal insights into the decisions of neural networks. We set gray occlusion region on the input image, and recorded the output probability while different regions were occluded. By converting these probability changes into heat map, the key parts of the input image for decision could be displayed.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.