Some research groups are working intensively on the automated evaluation of wound healing experiments. Recently, deep learning methods have also been evaluated for this application. Often, the focus is on tracking cells, generating cell paths and deriving information about cell interactions. For example, Ulicna et al. developed an analysis tool (DeepTree) consisting of two neural networks, one for segmentation (U-net) and the other for deciding the status of the cell, like mitosis and apoptosis [19]. Other groups focus more on segmentation accuracy by developing and establishing new network structures [20,21,22,23]. They derive metrics from cell experiments that can be evaluated using positional information from individual cells. But, information on population scale, particularly in wound healing, is lacking. Recently, Javer et al. published a different deep learning approach for analyzing scratch assays. They do not concentrate on the exact segmentation of cells but on the positional probability of the cell center for investigation of collective cell motion of scratch assays (DeepScratch).
We present a deep learning MATLAB application for investigation of live cell images on cell scale and population scale. For demonstration a wound healing experiment under different flow conditions (1.5 dyn vs. 10 dyn) was used. The workflow of the MATLAB application is shown in Fig. 1.
The MATLAB application consists not only of the U-net based iCD module but also of a training module. The training module starts with semi-automatic labeling of live cell images from wound healing assays. For this purpose, we created an algorithm with an adaptive thresholding method for segmenting cells and manually processing incorrect or inaccurate segmentations. This labeled data can be augmented to increase the training data and decrease the number of semi-manual cell labels needed. For creating the iCD network, we used a U-net architecture developed by Ronneberger et al. [24]. Using the trained network, we focus on segmenting each cell as accurately as possible. The segmented cell images were used for different application on cell scale (spatial distribution of cell density and cell velocity) and population scale (leading edge detection and wound closure analysis).
Cell detection and segmentation
Cell detection is the initial step for further analysis presented in this paper. Live cell images of wound closure were obtained every 15 min over a period of up to 10 h. For validation of the iCD training we calculated the intersection over union (IoU), and the Boundary overlap ratio (F1-score) using 42 cell images with at least five cells at each frame. Overall 1467 individual cells were manual segmented. The IoU and F1 was computed for each of the Images using the MATLAB function ‘jaccard’ and ‘bfscore’. The mean IoU reaches a value of 0.8214 ± 0.038 and the F1 score 0.9178 ± 0.045. Manually, an average of 34.93 cells were detected on the validation images, while the iCD method detected 32.83 cells on the validation images (p-value: 0.644). Therefore, there is no statistically significant difference in the number of cells counted.
The adaptive threshold method is based on the local grey value distribution. Therefore, we manually defined a certain region of interest as well as a threshold sensitivity, which means that this method cannot be considered user independent. An area of interest of 161 × 121 pixels and sensitivity 0.5–0.7 were suitable parameters. To assess differences between the described methods, we applied the different cell detection methods (iCD, adaptive threshold method and freehand detection) to calculate the initial cell density in a dynamic wound healing assay using human coronary artery endothelial cells (HCAECs, see Cell density section).
Cell density
We compared different cell detection methods (iCD, adaptive threshold method and freehand detection) regarding their suitability for cell density assessment. For cell density analysis each image was divided into 50 columns, analogous to Jin et al. [10]. The number of cells was counted in each column and the total number of cells, per column was divided by the column area to obtain the cell density for each column. For illustration of the quality of the different approaches Fig. 2 exemplarily presents detected cells of a cell culture experiment.
For adaptive threshold we observed an overall relative error in the total cell count of 1.6% and for our iCD approach 8.5%. The relative error refers to manual data.
To evaluate the temporal wound healing process, the column-averaged cell density, obtained by the adaptive threshold method and by the iCD approach, was plotted for each time point. Exemplarily Fig. 3 depicts the results obtained by iCD (t = 0 h up to 10 h, ∆t = 15 min).
The live cell images were of high quality (Fig. 2) regarding noisiness. Next, the robustness of these two methods against image distortion was tested by manipulating high-quality raw images applying the MATLAB Gaussian noise filter. The analysis of cell density was repeated by means of adaptive threshold method and iCD using the distorted images (Fig. 4).
In general, both methods showed similar cell density values and distribution when using high-quality raw images (Fig. 4a and c). The asymmetry of the gap closing is caused by the use of live cell images under flow conditions. By applying the MATLAB Gaussian noise filter, the cell density of the scratch obtained from the adaptive threshold method differs from the former results using the high-quality raw images (compare Fig. 4c and d). Especially in the region of the gap, where no cells are present, noisy images lead to erroneous cell detection. However, the results of the iCD method were barely influenced by image manipulation as shown in Fig. 4b compared to adaptive threshold (compare Fig. 4a).
Cell velocity
We compared different approaches to validate the detection of cell motion: adaptive cell image velocimetry (CIV), iCD-based cell tracking and manual tracking of individual cells (Fig. 5a). The latter was used as reference. Since the gap between the cell monolayers is positioned perpendicular to the x-direction, averaging characteristics in the y-direction (column-wise), such as the cell velocity, are assumed to be valid. Based on the velocity field of ECs column-wise averaging of the velocity magnitudes was performed, resulting in a velocity function, which only depends on the x-direction. As reference, the cell velocity was investigated three times by manual cell tracking for one time step (at time point t = 7.5 h, ∆t = 15 min); Fig. 5a). This time step was selected because the wound healing was well established and therefore all cells were set in motion, even at a distance from the gap.
We obtained an average velocity magnitude of 0.031 mm/h with an absolute and relative error between different manual analyses of 7 mm/h and 22%, respectively. The relative error refers to manual data. The average velocity detected by CIV was 0.012 mm/h with an averaged absolute and relative error of 0.020 mm/h and 62.7% compared to manual tracking. The average velocity detected by iCD was 0.036 mm/h with an averaged absolute and relative error of 0.0047 mm/h and 14.5%, compared to manual tracking.
To investigate the cell velocity during the healing process we applied the adaptive CIV and iCD method on every live cell image (t = 0 to t = 24 h). The results are plotted in Fig. 5b and c. A difference between the cell velocities resulting from iCD and CIV was also observed over time. For example, a maximum cell velocity of > 0.05 mm/h was determined using iCD, whereas the highest cell velocity values for CIV were 0.03 mm/h.
Leading edge detection and wound closure analysis
Leading edges were detected by the conventional Canny method (orange), by a second deep learning approach: called intelligent direct scratch detection (iDSD) (light blue), iCD (dark blue), and manual (green), see Fig. 6 (time point = 3.75 h). The leading edge obtained by Canny method and iDSD show a tendency to become more frayed, while the contour obtained from the iCD method is in better alignment to the freehand line.
A quantitative evaluation of the leading edge detection can be achieved by the highly sensitive edge length or so-called edge protrusion. Again, we compared the Canny method, our two different CNN approaches (iDSD and iCD) against the freehand edge as reference. The edge length was measured every 15 min over 6 h. In Fig. 7(a-d) we plotted the absolute edge length of both upstream and downstream edges. Manual data were obtained three times for each time frame. The average values at each time point is displayed in the diagram, with its corresponding standard deviation and is used as reference for the relative error computation. For statistical analysis the relative errors of leading edge protrusion are compiled in a boxplot shown below (Fig. 7e).
The relative error in the determining of the leading edge protrusion is lowest for freehand values, but even here a relative error of 5.9% is found on average. Referencing the freehand line, the iCD method has the lowest relative error regarding edge protrusion (11.7%) and has the smallest variation around the mean value. An average relative error for left and right edge protrusion of 76.4% were found using the Canny method and 42.4% using the iDSD approach. All methods detected an increasing edge length, which indicate the migration of pioneer cells into the wound. In general, the Canny method overestimated the edge length compared with freehand detection.
The wound healing process can also be quantified by the position of the leading edge as a function of time. Therefore, based on the former edge detection the spatially averaged edge position was calculated at different time steps (Fig. 8a, b) using Canny, iDSD, and iCD method in comparison to the freehand analysis (3 independent datasets were used). For statistical analysis the relative errors of spatially averaged edge positions are compiled in a boxplot shown below (Fig. 8c).
All methods detect a comparable wound healing behavior. It can be seen, that under low flow (1.5 Pa) the upstream edge is slightly faster than the downstream edge. The difference is more pronounced with increasing flow (10 Pa). Even if the Canny method is distorted by imaging artefacts, the edge position is in a very good agreement with the manual detection due to spatial averaging. The iDSD localized upstream edge behind the edge positions, which were obtained by other methods. The reason for this is shown in an example image, where the direct detection of the edge works very well in the beginning and can lead to false detection in the further time through changes of the image quality (Fig. 8). When determining the spatially averaged edge position, the iCD method shows the same mean relative error (2%) as freehand, but a slightly higher variation. Both the Canny method (2.6%) and the iDSD (4.2%) approach showed higher relative errors in the determination of the spatially averaged edge position.