Figure 2.
Example applications using MIA
(A–D) Neural networks were trained with MIA default settings for segmentation (A), classification (B), object detection (C), and tracking (D) using different datasets.
(A) A neural network was trained to segment individual cells from light microscopy using the LIVEcell dataset.27 The model achieved an intersection over union (IoU) of 0.8 ± 0.01 on the test data. Example images from the test data are shown in pairs, with the original image on the left and the same image with the model predictions overlaid in yellow on the right.
(B) A neural network was trained to classify lymph node tissue in normal (green) or containing metastatic tissue (red).25,26 The model achieved accuracy of 86% ± 2% to correctly classify metastatic tissue in the test data.
(C) A neural network trained for object detection achieved a mean absolute error (MAE) of 0.5 ± 0.01 on the identification of dividing (red) and non-dividing (green) cells expressing H2b-GFP.22
(D) Results from object detection or segmentation can be used to perform tracking of individual objects over consecutive time points. Each object is shown in an individual color and traces refer to positions along previous time points. Results are reported as the mean ± SD of 3 training runs.