A graph-based cell tracking algorithm with few manually tunable parameters and automated segmentation error correction

Automatic cell segmentation and tracking enables to gain quantitative insights into the processes driving cell migration. To investigate new data with minimal manual effort, cell tracking algorithms should be easy to apply and reduce manual curation time by providing automatic correction of segmentation errors. Current cell tracking algorithms, however, are either easy to apply to new data sets but lack automatic segmentation error correction, or have a vast set of parameters that needs either manual tuning or annotated data for parameter tuning. In this work, we propose a tracking algorithm with only few manually tunable parameters and automatic segmentation error correction. Moreover, no training data is needed. We compare the performance of our approach to three well-performing tracking algorithms from the Cell Tracking Challenge on data sets with simulated, degraded segmentation—including false negatives, over- and under-segmentation errors. Our tracking algorithm can correct false negatives, over- and under-segmentation errors as well as a mixture of the aforementioned segmentation errors. On data sets with under-segmentation errors or a mixture of segmentation errors our approach performs best. Moreover, without requiring additional manual tuning, our approach ranks several times in the top 3 on the 6th edition of the Cell Tracking Challenge.

A graph-based cell tracking algorithm with few manually tunable parameters and automated segmentation error correction

Automatic cell segmentation and tracking enables to gain quantitative insights into the processes driving cell migration. To investigate new data with minimal manual effort, cell tracking algorithms should be easy to apply and reduce manual curation time by providing automatic correction of segmentation errors. Current cell tracking algorithms, however, are either easy to apply to new data sets but lack automatic segmentation error correction, or have a vast set of parameters that needs either manual tuning or annotated data for parameter tuning. In this work, we propose a tracking algorithm with only few manually tunable parameters and automatic segmentation error correction. Moreover, no training data is needed. We investigate the performance of our approach by simulating erroneous segmentation data, including false negatives, over- and under-segmentation errors, on 2D and 3D cell data sets. We compare our approach against three well-performing tracking algorithms from the Cell Tracking Challenge. Our tracking algorithm can correct false negatives, over- and under-segmentation errors as well as a mixture of the aforementioned segmentation errors. Furthermore, in case of under-segmentation or a combination of segmentation errors our approach outperforms the other tracking approaches.

Stepwise segmentation error correction in optical coherence tomography angiography images of patients with diabetic macular edema

Aim: To determine the minimum number of optical coherence tomography B-scan corrections required to provide acceptable vessel density measurements on optical coherence tomography angiography images in eyes with diabetic macular edema. Methods: In this prospective, noninterventional case series, the optical coherence tomography angiography images of eyes with center-involving diabetic macular edema were assessed. Optical coherence tomography angiography imaging was performed using RTVue Avanti spectral-domain optical coherence tomography system with the AngioVue software (V.2017.1.0.151; Optovue, Fremont, CA, USA). Segmentation error was recorded and manually corrected in the inner retinal layers in the central foveal, 100th and 200th optical coherence tomography B-scans. The segmentation error correction was then continued until all optical coherence tomography B-scans in whole en face image were corrected. At each step, the manual correction of each optical coherence tomography B-scan was propagated to whole image. The vessel density and retinal thickness were recorded at baseline and after each optical coherence tomography B-scan correction. Results: A total of 36 eyes of 26 patients were included. To achieve full segmentation error correction in whole en face image, an average of 1.72 ± 1.81 and 5.57 ± 3.87 B-scans was corrected in inner plexiform layer and outer plexiform layer, respectively. The change in the vessel density measurements after complete segmentation error correction was statistically significant after inner plexiform layer correction. However, no statistically significant change in vessel density was found after manual correction of the outer plexiform layer. The vessel density measurements were statistically significantly different after single central foveal B-scan correction of inner plexiform layer compared with the baseline measurements ( p = 0.03); however, it remained unchanged after further segmentation corrections of inner plexiform layer. Conclusion: Multiple optical coherence tomography B-scans should be manually corrected to address segmentation error in whole images of en face optical coherence tomography angiography in eyes with diabetic macular edema. Correction of central foveal B-scan provides the most significant change in vessel density measurements in eyes with diabetic macular edema.

Effect of segmentation error correction on optical coherence tomography angiography measurements in healthy subjects and diabetic macular oedema

PurposeTo evaluate the impact of segmentation error on vessel density measurements in healthy eyes and eyes with diabetic macular oedema (DMO).MethodsIn this prospective, comparative, non-interventional study, enface optical coherence tomography angiography (OCTA) images of the macula from healthy eyes and eyes with DMO were acquired. Two expert graders assessed and corrected the segmentation error. The rate of segmentation error and the changes in vessel density and inner retinal thickness after correction of the segmentation error were recorded and compared between the two groups.Results20 eyes with DMO and 24 healthy eyes were evaluated. Intergrader agreement was excellent (intraclass correlation coefficient ≥0.9) for all parameters in both groups. The rate of segmentation error was 33% and 100% in healthy and diabetic eyes, respectively (p<0.001). Nine healthy eyes (37.5%) and all eyes with DMO (100%) were noted to exhibit a change in at least one of the foveal or parafoveal vessel density measurements. The rate of any change in foveal and parafoveal vessel densities in both the superficial and deep capillary plexus was statistically significantly higher in the diabetic group (all p<0.001). No statistically significant change was observed in mean vessel density (superficial and deep capillary plexuses) after correction of the segmentation error in healthy and DMO eyes (All p>0.05). However, the mean absolute change in the vessel density measurements was statistically significantly higher in the diabetic group (all p<0.05). The mean absolute change in superficial and deep inner retinal thickness was statistically significantly higher in DMO (p=0.02 and p=0.002, respectively).ConclusionsIn this study, misidentification of retinal layers and consequent vessel density measurement error occurred in all eyes with DMO and in one-third of healthy eyes. The segmentation error should be checked and manually corrected in the OCTA vessel density measurements, especially in the presence of macular oedema.

Application of convolution neural networks in eye fundus image analysis

In this work, we proposed a new approach to analyzing eye fundus images that relies upon the use of a convolutional neural network (CNN). The CNN architecture was constructed, followed by network learning on a balanced dataset composed of four classes of images, composed of thick and thin blood vessels, healthy areas, and exudate areas. The learning was conducted on 12x12 images because an experimental study showed them to be optimal for the purpose. The test error was no higher than 4% for all sizes of the samples. Segmentation of eye fundus images was performed using the CNN. Considering that exudates are a primary target of laser coagulation surgery, the segmentation error was calculated on the exudate class, amounting to 5%. In the course of this research, the HSL color system was found to be most informative, using which the segmentation error was reduced to 3%.