This work was partially supported by Ministerio de Economía y Competitividad (MTM2017-88804-P), and Agencia de Desarrollo Económico de La Rioja (2017-I-IDD-00018).SNR, PSNR, RMSE, MAE ImageJ's plugin to assess the quality of images We have tried other segmentation methods using K-means and Canny edge detection however, those methods produce worse results than those obtained using the previous methods. In and Iv stand for Insidia and Ivanov, respectively. The best result for each dataset is highlighted in bold face, *** significant difference between methods. Mean (and standard deviation) for the fluorescence datasets. Mean (and standard deviation) for the brightfield datasets. Our plugin has been tested with several datasets of images acquired under different experimental conditions. The best model was constructed using the HRNet-seg architecture. We have also trained several deep learning models with a set of notebooks and two datasets of spheroid images. The two images produced in this way are combined using the AND binary operation to output the mask. To this aim, the normal image is processed by sequentially finding its edges and binarising it and, the fluorescence image is binarised using the IsoData thresholding method. Finally, we have designed an algorithm that takes advantage of images acquired with fluorescence. This strategy applies both Algorithms A1 and A2 to the input image, adds the two resulting masks, and fills the holes of the resulting mask to produce the final output.Ī5. Namely, it starts applying the threshold approach, and if it fails to find a valid mask, it applies the edges approach.Ī4. This approach is a sequential application of Algorithms A1 and A2. The process stops after a valid mask is found or when a number of iterations is reached.Ī3. In case that the method does not work, the number of iterations that the find edges operation is applied is increased. The second strategy does not directly binarise the image but it firstly finds the edges of the image, and subsequently binarise the image using the IsoData method. This straightforward approach is useful when the spheroid image can be clearly distinguished from the background of the image.Ī2. In those cases where such a direct approach does not produce a valid mask, we sequentially binarise the image, dilate it, fill the holes, erode the image, and, finally, apply the watershed operation. The first strategy is based on just binarising the spheroid images by using the IsoData method. In addition, several variants of our algorithm are combined to deal with those cases where a proper spheroid mask is not generated.Ī1. Particular casesĭue to the different nature of spheroid images, we have particularised our generic algorithm using 5 strategies that is, using different values for the 5 parameters of our segmentation algorithm. Namely, the procedure can be split into two steps: contour generation and contour refinement.Īn example of the application of our generic algorithm is presented in the following image. This process, that is diagrammatically described in the following figure, is based on the sequential application of several image processing techniques, such as edge detection or thresholding, and morphological operations like dilation or erosion. Given an image containing a spheroid, our generic algorithm aims to produce a mask for the region that contains such a spheroid. A graphical interface built on top of ImageJ to employ SpheroidJ can be downloaded from here.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |