A novel method for evaluating and visualizing scratch wound healing assays using level-set and image sector analysis

Abstract

Scratch wound healing assays are widely used to study collective cell migration, essential for understanding tissue regeneration, drug effects, and wound healing mechanisms. However, conventional analyses often rely on wound edge dynamics or individual cell tracks, limiting spatial insight into migration behavior. We present a sector-based analytical framework that reinterprets time-lapse microscopy data by dividing each image into defined spatial regions across the field of view. This enables spatially resolved characterization of how cell populations migrate over time. To address challenges of low contrast and uneven illumination in bright-field microscopy, we apply a level-set segmentation algorithm that robustly detects the wound edge. Using this approach, we show that both cell velocity and trajectory vary with distance from the wound boundary. We introduce a novel metric, the sector–boundary distance, to identify regions where cells migrate faster along nonradial paths. To assess chemotactic activation, cells were treated with the chemokine CXCL10 to stimulate motility via CXCR3-mediated signaling. Statistical testing showed that, in treated cells, the proportion of highly motile cells was significantly associated with wound closure, even in regions distant from the scratch, whereas directionality played a limited role. By contrast, untreated cells exhibited weaker and less organized migration patterns. These findings highlight how local cellular activity contributes to healing in a treatment-dependent manner. Our method bridges global wound-level analysis and local cell-scale behavior by combining single-cell tracking with precise boundary detection. The complete framework is available as open-source software, including a user-friendly web application that enables interactive analysis of microscopy data.

Significance Statement.

Scratch wound healing assays are widely used to study collective cell migration, yet standard analytical approaches often overlook how migration dynamics vary across the wound area. Here, we introduce a sector-based framework that combines robust wound edge detection with spatiotemporal analysis of cell trajectories in time-lapse microscopy. By defining a parameter termed sector–boundary distance, we show that migration speed and direction depend on proximity to the wound edge, revealing nonradial patterns not captured by traditional methods. This approach connects global wound closure metrics with localized cell behavior, supporting a clearer understanding of how cells respond to spatial context during regeneration. The accompanying tool enables reproducible analysis and offers a practical extension to established wound healing assays.

Introduction

The scratch wound healing assay was developed as a method to study cell migration and the mechanisms underlying wound healing (1). It is now a well-established, simple-to-perform, versatile, and inexpensive assay, which has been used to investigate a wide range of biological questions. With considerable advancements in live cell imaging and computational analysis methods, it is a widely used assay in cell biology (2, 3).

The original form of the assay is performed by creating a scratch in a confluent monolayer of cells, followed by measuring the rate at which cells migrate to the scratch using time-lapse microscopy (4). The basic idea is to track the movement of the leading edge of the scratch (a term used for a horizontal or vertical void space in a cell monolayer with a consistent width along its length) or wound (sometimes called a pseudowound; this term applies to both scratches and spaces of other shapes, such as circles) over time. This tracking allows evaluation of the effects of different biologically active components on the migratory ability of various cell types (5, 6).

The scratch wound healing assay has become a widely used method not only for the study of cell migration but also for evaluating the regenerative potential of bioactive compounds, especially in the context of wound healing. It has been frequently applied in drug development, particularly as a primary screening tool to identify compounds that either promote or inhibit cellular migration (7, 8). For example, Scrima et al. (7) demonstrated that an extract from Salvia haenkei significantly accelerated the migration of both keratinocytes and fibroblasts in a scratch assay and also enhanced re-epithelialization in an in vivo murine wound model. Similarly, Li et al. (8) showed that overexpression of microRNA-132 in dermal fibroblasts led to significantly faster wound closure in vitro, with RASA1 identified as a key downstream target.

Beyond testing proregenerative substances, the assay is utilized to investigate pathological conditions associated with aberrant fibroblast behavior, such as keloid formation. Alishahedani et al. (9) found that keloid-derived fibroblasts displayed accelerated closure rates in scratch assays, along with overexpression of inflammatory cytokines, compared with normal fibroblasts. Treatment with known bioactive compounds such as caffeine and allicin successfully attenuated this excessive migratory behavior, indicating their potential role in topical antikeloid therapy.

Although the assay is traditionally conducted in vitro, these applications are often part of integrated experimental pipelines in which promising compounds are further tested in in vivo wound models (7). The assay thereby plays an essential role in preclinical wound healing studies, providing a cost-effective system for identifying promising candidates for more complex in vivo validation methods.

Fibroblasts are central to the animal wound healing process, utilizing proliferation, remodeling, and migration into the wound bed, essential for extracellular matrix deposition and wound contraction. However, accumulating evidence suggests that the mode of fibroblast migration also influences the quality of tissue regeneration and the extent of scar formation. Thus, understanding fibroblast migration is crucial for evaluating tissue regeneration, drug effects, and wound healing mechanisms. Scratch wound healing assays are widely used to study collective fibroblast behavior; conventional analysis methods for the data generated by this assay focus mainly on evaluating the closure of the scratch edge or single-cell migration (10). However, there is a growing body of evidence indicating that the coordinated movement of groups of cells, termed collective cell migration, is critical for wound healing (11–13). Figure 1 illustrates how a scratch wound healing assay using time-lapse microscopy is performed.

Fig. 1.

Illustration of a scratch wound healing assay conducted with a device designed for time-lapse microscopy of living cells. In this experiment, cells are cultured in a plastic well plate, which typically consists of multiple wells depending on the experimental setup. After adding the necessary chemicals, a scratch is introduced using a pipette tip. Although not the most precise tool, the pipette tip remains a common choice for this procedure. The well plate is then placed in a time-lapse microscopy device, where each well is imaged at predefined intervals. The resulting data are analyzed using manual or automated image processing techniques to evaluate wound closure.

Motivation and characteristics of the issues addressed

When evaluating the migration behavior of chondrocytes and fibroblast-like synoviocytes using scratch assays, we encountered several issues that needed to be addressed to assess the experiments accurately. The first issue involved the analysis of microscopic images. In bright-field microscopy, images were negatively affected by low contrast and uneven brightness distribution and were further compromised by the deliberate inclusion of synovial fluid in the growth medium, which introduced nonliving contaminants and reduced overall image quality (14). Consequently, the leading edge of the scratch was difficult to detect (compared with other published work, where the transition between the cell monolayer and the scratch is more contrasted (4, 15, 16).)

The second issue arises from the biological and physical characteristics of fibroblast and fibroblast-like cell movement on a smooth surface within the cell monolayer. Fibroblasts that migrate to the scratch tend to form a large lamellipodium extending to the scratch, accompanied by a reduction in stress fibers within the cell (17). This behavior leads to irregular movements along the scratch border and the formation of difficult-to-detect ingrowths (18). Furthermore, these movements are influenced by the process of scratch formation within the cell monolayer (19), which can introduce mechanical artifacts and alter the local migration dynamics. These irregularities, along with the variability introduced during scratch creation, make it particularly challenging to evaluate migration using traditional scratch wound healing assay analysis methods. Such methods are typically designed for more uniform and collective migration patterns and may not be well suited to capturing the complex and heterogeneous behavior of fibroblast-like cells.

The most notable challenge we encountered lay in the inherent limitations of this assay when applied to fibroblasts and fibroblast-like cells, which are a consequence of the natural collective migration in the cell monolayer (17, 20). Bright-field microscopy tracking of the leading edge provides only basic information, such as the speed at which the two edges move toward each other. With the capabilities of fluorescence microscopy and rapid automated detection of cell nuclei using deep learning methods, it is possible to improve the evaluation of scratch assays by incorporating additional information on the specific movement of cell communities around the scratch (3, 15).

Brief characterization of cell migration during a scratch wound healing assay

The migratory behavior of cells during a scratch wound healing assay can be interpreted with two types of cell movement: individual migration of single cells and coordinated migration of a group of cells, resulting in collective cell migration (21). The correct evaluation of each type of migratory behavior requires different approaches, both in the experimental modeling and in the methods used to assess their overall contribution (22).

The interplay between the movement of individual cells and the migration of the collective is usually not distinguishable in traditional analysis methods despite its vital role in wound healing, as well as other biological processes such as cancer progression (23). However, in a nonconfluent scratch assay, the interaction between leader and follower cells may be more complicated (21), and the correct evaluation of collective cell migration is therefore more challenging. Moreover, cells from the monolayer that are not on the edge of the scratch can migrate to the edge to become leader cells (24, 25). A critical aspect of leader and follower cells is the effect and number of leader cells with respect to the displacement and direction of the collective migration of the follower cells. Direct intercellular physical interaction, the pulling force, or signaling between these cells makes this movement possible (26). By contrast, in scratch assays, several cells on the edge of the scratch will effectively direct the migration direction of other cells, with the interaction between these leader cells also playing a role in the effective closure of the scratch. Based on the topology of migrating populations, this “leader-follower” behavior can contribute to the persistent collective directional movement of cells to close the scratch (27).

Approaches to quantifying cell migration in scratch assays

Two main strategies are commonly used to evaluate scratch wound healing assays: (i) measuring changes in wound geometry over time and (ii) tracking individual cells and their contribution to wound closure (3). Both require accurate segmentation to distinguish between the cell monolayer and the wound area (Fig. 2). The scratch closure rate can be assessed by monitoring wound area reduction or the decrease in average scratch width (4, 16), with closure velocity offering an advantage by being independent of the initial wound size (28). These geometric approaches are especially suited to cell types exhibiting sheet migration characteristics (explained below) (23, 29–31).

Fig. 2.

A tracking sequence of images captured during a scratch wound healing assay. The images depict the monitoring and categorization of cells based on their contribution to the wound healing process, with colored cells indicating cells that contribute to the wound, level-set-detected boundaries and fluorescent imaging; A is the area of the scratch, which is calculated as the total sum of pixels between the right and left edges of the scratch as detected by the level-set and expressed in square micrometers. The top row (a, b, c) exhibits the boundaries detected by the level-set in bright-field images and visualized in the images in the fluorescence spectrum. The second row (d, e, f) highlights the 200 cells that migrate the most, determined by the distance traveled.

Alternatively, the second strategy, tracking fluorescently labeled nuclei, allows the number of cells that populate the wound over time to be quantified (32), with some models using the Fisher–Kolmogorov (reaction–diffusion) model, which describes the wound closure process as the development of cell density in space and time due to two mechanisms, diffusion and proliferation, both of which are potentially involved in the wound closure process in most tissues (33–36).

Cell density plays a critical role in migration dynamics and scratch wound healing assay reproducibility (37, 38). Its impact has led to modeling approaches that incorporate transport phenomena and initial density differences (15). Figure 3 illustrates how bright-field, fluorescence, and fused microscopy data can provide complementary insights into wound shape, cell density, directionality, and ingrowth patterns.

Fig. 3.

Information provided by different types of microscopy in scratch wound healing assays. It includes (i) bright-field microscopy images highlighting scratch/wound morphology, (ii) fluorescent images of stained cell nuclei and (iii) combined images integrating both methods. These combined images enable the definition of key parameters, such as the distance of cells from the advancing wound edge, as well as the quantity, speed, and prevailing direction of cells within individual image sectors. Additionally, the specific morphology of the cells determines the characteristic shape of the wound, which is influenced by the ingrowth of these cells into the wound area.

Importantly, the choice of evaluation strategy depends on the specific cell type and migration mode. Epithelial and endothelial cells often migrate as cohesive sheets, consistent with the wave-like models predicted by the Fisher equation (34, 39). By contrast, fibroblasts and related mesenchymal cell types display diverse migratory behaviors shaped by their role in wound repair, including polarization, cytoskeletal remodeling, and matrix interaction (40, 41). In 2D assays, they migrate as loosely interacting populations with partial sheet-like features (42). Our proposed method accounts for both individual cell activity and community-level behavior, offering a more comprehensive view of fibroblast-like cell migration during wound healing.

Results

Our results are intended to demonstrate that implementing a level-set algorithm together with the subsequent use of outputs from available and commonly used software is a method for performing scratch wound healing assays suitable for microscopic images possessing generally low and cross-image variable contrast. A total of four positions (regions of interest [ROIs]) per well (one experiment) were analyzed, with each position having different image characteristics. The sector-based analysis focuses on analyzing the behavior of a community regardless of the original positional information and speed of the cells. The sector-based analysis method is compatible with any wound healing image analysis workflow and, without requiring additional experimental steps, improves the robustness of cell migration assessment. Using this method in combination with single-cell tracking allows for a 2-fold analysis of the same images, with the addition of sectors enabling cell behavior to be followed based only on their sector–boundary distance, which is an effective range for the cells on the edges to influence the cells in each box. For a visualization of the wound healing assay based on the sector method, refer to Fig. 4.

Fig. 4.

Visualization of the wound healing assay involves application of the newly developed sector method on three regions of interest within a single sample. The first row demonstrates the use of the level-set algorithm with bright-field images, the second row exhibits segmented nuclei in fluorescent images, and the third row visualizes the outcomes achieved through the newly developed sector method.

The average direction of movement of the cells in each sector showed a significant correlation with the distance of each sector from the scratch boundary (the sector–boundary distance). The difference between the number of top-speed cells and the total number of cells in each sector was significantly different between sectors that were closer to the scratch compared with those farther away by more than $300\mu \mathrm{m}$. Consequently, the average velocity of cells in the sectors that were closer than $300\mu \mathrm{m}$ to the wound was greater (Mann–Whitney U test, $P<0.00001$) (Fig. 5). Furthermore, the average cell velocity in the x-axis direction at a distance close to the wound (distance of $300\mu \mathrm{m}$) was higher in the first half of the experiment (Mann–Whitney U test, $P<0.00001$) (Fig. 6). As the rate of scratch retraction is initially higher than at later time intervals when the scratch is already partially closed, the rate of scratch retraction represents another parameter that can be evaluated with this method and can be compared under different conditions.

Fig. 5.

a) Distribution of the average velocity in the x-axis direction of the cells based on the distance from the wound boundary and the prevailing direction; the vertical line $x=300\mu \mathrm{m}$ separates the sectors closer to the wound boundary, i.e. the sectors with a higher abundance of top-speed cells. This figure provides a comprehensive depiction of cellular dynamics by presenting information on both speed and movement patterns for all observed sectors of the ROI. This information enables differentiation of sectors exhibiting varying degrees of cellular motion activity, ranging from slow to top-speed cells. The occurrence of top-speed cells in the sectors further from the scratch boundary is caused by irregularities in the distribution of cells in the monolayer, which is why approximately the same number of cells in these sectors move toward and away from the wound. By contrast, (b), representing the conventional approach results, demonstrates a noticeable reduction in the provided information, thereby omitting crucial details that could contribute to a more nuanced understanding of cellular behavior.

Fig. 6.

Comparison of the average velocities of cells closer ($<300\mu \mathrm{m}$, purple) and further away ($>300\mu \mathrm{m}$, orange) from the scratch boundary in individual timesteps (1 timestep $=$ 20 min). The chart on the left (a) shows the average velocities in the x-axis direction; chart (b) shows the average resulting velocities (sum of the velocities in the x-axis and y-axis direction) in individual timesteps.

The average velocity of the cells and their direction, together with the direction and displacement of the top-speed cells, can link the two biological mechanisms of cell migration: single-cell displacement and collective migration. This type of analysis is more informative than the traditional analysis of cell migration in a wound healing experiment (as depicted in Fig. 7) because it is sensitive to differences in the width of the scratch in each sector while not being influenced by the noise of random cell movement. In addition, it matches the biology of the wound healing process more closely.

Fig. 7.

Evaluation of scratch wound healing assays using a traditional approach based on scratch width reduction over time, represented by boxplots of the scratch width reduction over time: a) data obtained by pooling four regions of interest (ROIs) within a single sample; b) boxplots capturing the variability among ROIs originating from the same sample, providing valuable insights into the diverse characteristics within the sampled data (1 timestep $=$ 20 min). This traditional approach illustrates the variability among ROIs within a single scratch wound healing assay. A well containing a 2D monolayer of cells was scratched using a pipette tip, validating the pooling of data from images within a single well.

Although the new method provides the same basic measures as the currently established methods, it makes it possible to have a more detailed view than in previous approaches. In addition, it allows for a more detailed view of the migratory behavior of cell populations by defining three new positional parameters: the distance of sectors from the advancing scratch boundary, the ratio of the number of top-speed cells to the total number of cells (active ratio) in each sector of the image, and the average velocity vector of cells in a sector over a defined number of time-lapse microscopy images. These parameters make it possible to quantify the type of contribution of a predefined cell community to wound closure, as well as retaining their positional information.

To evaluate the added value of our sector-based analysis, we directly compared it with the conventional approach commonly used in scratch wound healing assays. The traditional method quantifies wound closure by tracking the average scratch width over time and statistically comparing treatment conditions. Although informative at the global level, this approach overlooks the underlying cellular dynamics and spatial heterogeneity of cell behavior across the field of view. To illustrate this, we applied standard regression-based analysis to untreated (A1: CXCL10-negative) and treated (A2: CXCL10-positive) cell populations by fitting a linear model of the form $\mathrm{scratch}\mathrm{\_}\mathrm{width}\sim \mathrm{time}\times \mathrm{condition}$. This revealed a significantly faster decrease in scratch width in A2 than in A1, with estimated slopes of $-$28.40 and $-13.82\mu \mathrm{m}$/h, respectively. The interaction term was highly significant ($P<0.0001$), confirming accelerated wound closure in the treated condition (Fig. 8).

Fig. 8.

Comparison of wound closure dynamics between untreated (A1) and treated (A2) conditions based on linear regression of the average scratch width over time. The treated condition (A2) shows significantly faster closure, with a steeper slope and greater model fit, confirming the effect of CXCL10 stimulation.

However, to investigate the spatial dynamics driving these differences, we developed a complementary method based on cell-level features extracted from localized sectors of the image. Specifically, we analyzed only those sectors located more than $400\mu \mathrm{m}$ away from the wound edge, thereby focusing on remote cell populations whose contribution to wound closure cannot be captured by area-based metrics alone. For each sector, we computed the final change in scratch width (delta_scratch) and modeled it as a function of two explanatory variables: the absolute x-directional velocity of cells (abs_avg_vec_x), and the proportion of active (“top-speed”) cells within the sector (active_ratio).

Separate ordinary least squares (OLS) models were fitted for A1 and A2 using delta_scratch as the outcome variable, implemented in Statsmodels Python library (43). In A2, only active_ratio emerged as a significant predictor of scratch closure (coefficient $=$ 1,590.09, $P<0.0001$), whereas abs_avg_vec_x was not statistically significant (coefficient $=6.35,P=0.378$). The overall model was statistically significant ($F(2,2747)=28.67$, $P<0.0001$) but explained only a small portion of the variance ($R^2=0.020$).

By contrast, the A1 model showed only a weak effect—abs_avg_vec_x was marginally nonsignificant ($P=0.067$), whereas active_ratio remained significant ($P<0.0001$) but with much lower explanatory power ($R^2=0.008$, $F(2,2033)=8.56$, $P<0.001$).

These findings indicate that, in treated cells, a higher proportion of highly motile cells contributes to wound closure even from distal regions, despite the absence of a pronounced directional migration component. By contrast, the response in untreated cells appears less organized and less effective. Most importantly, the proposed approach offers spatially resolved, cell-level insight into the mechanisms of wound healing that remain inaccessible to conventional analyses based solely on average scratch width. Together with the sector-based maps in Fig. 9a–d, this framework couples imaging with quantitative modeling to reveal localized cellular contributions that are otherwise obscured in spatially aggregated measurements.

Fig. 9.

Sector-based visualization of active ratio and migration vectors. a–d) Spatial heatmaps of the active cell ratio with overlaid directional migration vectors for untreated and treated conditions at $t=2\mathrm{h}$ and $t=20\mathrm{h}$: a) untreated, $2\mathrm{h}$; b) treated, $2\mathrm{h}$; c) untreated, $20\mathrm{h}$; d) treated, $20\mathrm{h}$ are shown. This view emphasizes local, sector-level differences in cell behavior over time, beyond global motility metrics.

Discussion

Wound healing involves a complex interplay of processes, including inflammation, migration, and matrix remodeling. However, during an in vitro scratch wound healing assay, several of these processes, particularly hemostasis and the involvement of nonepithelial cell types, are absent. Moreover, this assay is usually performed with a limited type of cells with relatively fast migratory behavior, such as fibroblasts or established cell lines. Despite its popularity, the assay is impacted by reproducibility and interpretation issues due to variability in scratch creation, debris, or cell senescence, which can obscure true biological effects (44, 45). Standard metrics such as average scratch width fail to capture complex behaviors, including leapfrogging or stop-and-go motion, which are characteristics of fibroblast migration.

To overcome these limitations, recent efforts have focused on more robust image-based analyses. Our sector-based approach extends this by resolving cell behavior at defined distances from the wound. Applied to treated and untreated fibroblast-like cells, it revealed that treatment induced globally elevated but spatially uncoordinated migratory activity, even in distal regions—patterns that remain undetectable using conventional measures. By contrast, untreated cells exhibit less organized and less effective migration. These findings would remain undetectable using conventional approaches based solely on average scratch width. Our method allows the identification of heterogeneous migratory patterns, offering new opportunities to assess therapeutic interventions and cell-type-specific mechanisms in wound healing.

Several recent studies underscore the biological relevance of cell migration heterogeneity. For example, Jiang et al. (46) reported a subpopulation of fascial fibroblasts undergoing coordinated swarming migration during murine wound repair, contributing to fibrosis through N-cadherin-mediated adhesion. By contrast, fibroblasts derived from tissues that heal with minimal scarring (e.g. fetal skin or oral mucosa) migrate in a more dispersed and individual manner (47). These results reinforce the notion that fibroblast migratory behavior is highly context dependent, and our method offers a practical, scalable framework to detect and analyze such differences within standard assay conditions. In addition to swarming, fibroblasts display other coordinated migratory behaviors. In cancer, they can guide collective invasion by creating paths for tumor cell clusters, termed fibroblast-led collective invasion (48). In wound healing, fibroblasts often migrate along fibrillar structures such as collagen or fibrin in linear chains, guided by durotaxis (49). Unlike epithelial cells, which migrate as cohesive sheets, fibroblasts move individually or in small clusters, often rearranging their positions in a process known as shuffling (50). Although leapfrogging is mainly associated with epithelial cells (51), similar overtaking dynamics may occur in crowded fibroblast cultures. Vortex-like motions have also been observed in fibroblast monolayers, albeit in a less ordered fashion than in epithelia. These rotational patterns likely emerge from mechanical feedback between cells and the substrate and may reflect localized imbalances in tension, adhesion, or migratory coordination (52, 53). Notably, our sector-based method enables the detection of such distinct migratory features by quantifying directional persistence, the proportion of highly motile cells, and their distance from the wound edge, which are parameters that are otherwise inaccessible through conventional approaches.

Table 1 summarizes the main advantages of the proposed method.

Table 1.

Summary of the main advantages of the proposed method.

(a)	Considers the number of cells in each image sector and their change over time. This allows inclusion of cell density, which can be normalized.
(b)	Enables analysis of cell behavior at a specific distance from the wound boundary, suitable for populations that do not exhibit sheet migration (e.g. fibroblast-like cells).
(c)	Allows detection of zones where cells begin migrating faster and more linearly.
(d)	Identifies a threshold distance from the wound at which cells are activated to move more rapidly and directionally, supporting the study of biochemical responses.
(e)	Evaluates migration parameters at defined distances from the wound independently of specific time points, beyond just tracking wound closure speed.

Segmenting the wound edge in bright-field time-lapse images proved challenging due to variability in image quality and the dynamic morphology of migrating fibroblasts. To address this, we applied a level-set method, which robustly handled differences in contrast and shape complexity across samples.

Our approach minimizes operator bias and is resistant to sample-specific differences in cell density. Unlike standard methods, it combines cell velocity, directionality, and wound area reduction for more precise analysis. The main achievement of the method is its ability to define the contribution of each sector to migration based on local cell density and distance from the moving wound edge. This allows us to measure how the scratch position influences migratory behavior within each sector. Additionally, the method enables sector-wise computation of displacement and velocity.

Technical description

Our newly developed method builds upon established approaches by combining the tracking of the wound boundary with the monitoring of individual cell trajectories over time. Although initially designed to evaluate scratch wound healing assays, this method can also be extended to analyze other types of cell migration experiments. The main evaluation strategies for scratch assays, together with the novel sector-based approach with the ability to link local migration dynamics to the distance from the retracting wound boundary, are illustrated in Fig. 10.

Fig. 10.

While traditional evaluation of the scratch wound healing assay relies on tracking the wound boundary, our approach considers the contribution of both individual cell populations and cell communities during the process. The distance from the center of the sector to the wound edge $(\mathrm{sbd})$ is referred to as the sector–boundary distance; this distance allows the monitoring of the behavior of cell populations located further from the wound. The activation of so-called top-speed cells and their influence on overall wound closure is one of the parameters that can characterize the behavior of the specific type of cell under investigation.

Time-lapse image sequences were acquired using the Cytation 5 imaging system, optimized for live-cell microscopy. Each experiment, corresponding to one well of a microtiter plate, consisted of 60 time-lapse frames captured in both bright-field and dark-field modes. Image analysis was first performed to extract key spatiotemporal features, including the position of the wound boundary and the centroids of cell nuclei at each time point. These data were then integrated to derive novel parameters that quantitatively describe cell migration patterns during scratch wound healing assays.

Sample preparation and image acquisition

Image data for this study were obtained from the human synovial sarcoma cell line SW982. The defrosted cells were cultured in DMEM/F-12 with the addition of 10% fetal bovine serum (Thermo Fisher Scientific) and 1% Pen–Strep antibiotics (Merck). The cells were plated into a 12-well plate and incubated in 5% CO${ }_2$ at 37 ${ }^{\circ }$C to 80% confluence. The scratch was made using a $100\mu \mathrm{m}$ sterile pipette tip, and the nuclei of 6 of the 12 wells were stained with 5 $\mu \mathrm{L}$ of Hoechst 3,334 fluorescence dye (Merck). The images were taken every 20 min for 24 h by the BioTek Cytation5 Cell Imaging Multi-Mode Reader (BioTek Instruments, Blackfly BFLY-U3-23S6M camera), and four ROIs from each well were tracked using the following imaging properties: objective 4× (Olympus Plan Fluorite, NA: 0.13, working distance 17 mm) and image size 1,224 × 904 pixels. The experiment followed well-established protocols for scratch wound healing assays (54). To assess chemotactic activation, cells were treated with 100 ng/mL of CXCL10 for 60 minutes (Peprotech 300-12-25, Thermo Fisher Scientific Inc.) prior to the migration assay. This treatment was used to stimulate cell motility via CXCR3-mediated signaling.

Level-set method calculation for bright-field microscopic image segmentation

Chan and Vese’s original method was used for cell monolayer segmentation. This method was chosen because it is highly robust to noise and is able to segment even difficult-to-distinguish objects (55, 56). Rather than presenting all the details, we describe only the basic ideas. The calculation is provided in the Supporting Information.

Cell nuclei detection and tracking

Individual cell nuclei visible within the fluorescence spectrum were identified using the StarDist detector implemented in Fiji/ImageJ, which performs instance image segmentation (57). The algorithm uses a lightweight neural network based on U-net and works similarly to most detection algorithms. However, instead of axis-aligned bounding boxes, it predicts a star-convex polygon for each pixel to represent the shapes. To ensure that the time-lapse images were aligned accurately, the ImageJ plugin Image Stabilizer was employed (58). The automated tracking software TrackMate 7, integrated with a simple LAP tracker, was used to track cell nuclei (59, 60).

Data fusion and “sector” implementation

The following text describes the fusion of data obtained from wound edge tracking and cell nuclei tracking. The first step was to identify the magnitude of the motion of the cell nucleus most relevant to our goal, specifically determining the property of cell movement that contributes to wound closure. For our analysis, we focused on cells with the greatest track displacement. This metric measures the straight-line distance between a cell’s initial and final positions. Even if a cell makes a large excursion but returns to its starting point, the track displacement remains zero (60). We then determined the so-called top-speed cells, that is, the cells with the largest path shift, which also have high track mean speed and low mean directional change, which measures the angle between two succeeding links, averaged over all the links of a track. The data obtained by tracking nuclei from fluorescence images and the data obtained by implementing the level-set algorithm in bright-field images were merged. At this stage, the cell nuclei are already considered as spots—objects for which the shape of the contour is not important but for which the main information is provided by the $i,j$ positions (polygon centers detected by StarDist) over time—and their linking to form the “tracks.” Position $(S)$ of spot n is the following:

$$S_n(f,t)=(i_n(f,t),j_n(f,t)),$$ (1)

where $S_n(f,t)$ represents the $x,y$ coordinate of a spot n belonging to track t in a frame f. Due to the low average cell velocity, every fifth frame of the time series was used for subsequent data fusion.

In general, the level-set function $\varphi :\mathrm{\Omega }\to \mathbb{R}$ determines the segmentation of the image into the object (cell monolayer) and background (scratch). We define

$$\begin{array}{rl}{R}_1& =\{(i,j)\in \mathrm{\Omega }:\varphi (i,j)>0\},\\ R_2& =\{(i,j)\in \mathrm{\Omega }:\varphi (i,j)<0\},\end{array}$$ (2)

so that $R_1$ and $R_2$ are disjoint subsets of the domain (up to the zero level set). The wound boundary B is then given by the zero level set of ϕ,

$$B=\{(i,j)\in \mathrm{\Omega }:\varphi (i,j)=0\},$$ (3)

and, for each image row $r_j$, the left and right wound edges $B_1,B_2$ are taken from the boundary points along that row according to their horizontal positions. More details are provided in the Supplementary materials.

After data fusion, each frame (now containing the spots and wound boundary coordinates) is divided into 20 × 24 rectangles, each 61 × 37 pixels in size. The final dimensions were determined empirically by testing multiple sector sizes and evaluating their ability to capture meaningful spatial patterns of migration typical of fibroblast-like cell movement, while avoiding sectors that contain significant cell-free zones. The selected size also accounts for the range of intercellular signaling, which is limited to several hundred micrometers. Importantly, the sector size is not fixed and can be flexibly adjusted to accommodate different cell types, cell sizes, and migration dynamics.

Each sector S of an image is defined via the center coordinate $C_S=(C_x,C_y)$ and a frame f and results in the following data: the total number of cells, the number of top-speed cells, the average velocity vector of the cells in each sector in x- and y-directions, the average angles of the x- and y-velocity components ${\overline{\theta }}_x$ and ${\overline{\theta }}_y$ (in radians), time and the distance of the center of the sector from the scratch boundary sbd acquired from the level-set method, which is called the sector–boundary distance. The top-speed cells were filtered based on the largest change in position over time, that is, the first 100 cells in each ROI with the highest value of track displacement in each time frame.

The sector–boundary distance is defined for each sector S as

$$sb{d}_S={\displaystyle \underset{k\in 1,2}{min}}|C_S-B_k|,$$ (4)

for $B_k$ belonging to the row equal to the center coordinate $C_y$ of sector S. In other words, $sb{d}_S$ is the distance from the sector center to the closest boundary in the appropriate row.

These top-speed cells were also identified as those located in the outlier region of the cell velocity distribution. These cells may be the most aggressive in migration or possess unique properties that merit further investigation.

Data filtering involved removing the values of the width of the scratch in each sector that exceeded the width of the scratch at the initial time point $t_0$ (frame $f_0$) since the scratch width is expected to decrease with time. In addition, in the initial image, the scratch boundary was relatively uniform and easy to detect by the implemented level-set algorithm. This filtering method was used to eliminate outlier observations introduced by the algorithm’s performance at later time points.

Statistical analysis

All statistical analyses were performed in Python 3.10 (61) using the pandas (v2.2.2) (62), statsmodels (v0.14.0) (43), and NumPy (v1.26.4) (63) libraries. Linear regression models were fitted using the OLS method, as implemented in the statsmodels.formula.api interface. For time-series evaluation of scratch closure, interaction terms were included in the models to test differences in the slope between experimental conditions (scratch_width ∼ time × condition). In the spatially resolved analysis, we modeled the final scratch closure per sector (delta_scratch) using predictors such as the absolute value of x-directional cell movement (abs_avg_vec_x) and the proportion of motile cells in the sector (active_ratio). Statistical significance was evaluated using F-tests for model terms and t-tests for individual coefficients. Model assumptions were assessed using residual analysis, including visual inspection and formal normality tests (Omnibus and Jarque–Bera). A significance threshold of $\alpha =0.05$ was used throughout.

Implementation and availability

The complete source code for the image processing pipeline and the method described in this paper, along with a sample dataset, is publicly available on the Code Ocean platform.^a The capsule allows users to execute the code directly in a reproducible computational environment and explore its functionality using the provided dataset. Additional datasets are available upon request from the authors.

For more accessible and user-friendly interaction, we have also developed a dedicated web application. This application enables automated execution of the pipeline described in the Methods section on user-uploaded microscopy images. It supports visualization of wound closure over time and facilitates analysis of the directional movement of individual cell populations relative to the wound edge. The interface displays time-lapse sequences from both bright-field and fluorescence microscopy and presents the results through interactive visualizations and summary statistics. A screenshot of the user interface is shown in Fig. 11. The source code of the web-based visualization tool is openly available on GitHub.^b

Fig. 11.

User interface of the web application for automated scratch healing analysis. The central panel displays a microscopy image with the detected wound edge highlighted using a level-set method, with arrows indicating the dominant movement directions of cell populations. The left-side menu allows the user to select specific regions of interest within the image. On the right, a dynamic table presents calculated metrics and statistical summaries for either the selected region or the entire image.

Conclusion

By defining sectors within the image and combining the traditional approach to wound healing assay evaluation, scratch boundary tracking, with a more modern deep learning-based cell nucleus tracking approach, we can achieve a more comprehensive evaluation of time-lapse microscopy image data. This combined method considers both the movement of the scratch leading edge and the migration of cells within each sector of the image. This approach allows for a better understanding of how cells move in response to the scratch, highlighting the region characterized by rapidly moving cells migrating toward the scratch, and how this behavior changes over time.

The relevance of this overlay method to cell migration biology lies in its ability to differentiate the contributions of various migration mechanisms. Specifically, this method identifies whether the acceleration of cell movement in a particular direction is driven by the overall increase in speed and directionality of individual cells or the activation of a subset of fast-moving leading cells that then guide the migration of the rest. In the context of the scratch wound healing assay, this approach allows for the possibility that both mechanisms may be at play.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

This work was supported by an internal grant project of VSB - Technical University of Ostrava (SGS projects, grant number SP2025/018), in part by the Grant Agency of the Ministry of Health of the Czech Republic, Czechia (NW24-10-00395) and MH CZ—DRO (FNOl, 00098892).

Author Contributions

M.V. contributed to conceptualization, investigation, methodology, visualization, and original draft preparation. M.K. was responsible for methodology, software development, data curation, and original draft preparation. A.G. contributed to conceptualization, validation, investigation, and original draft preparation. O.M. contributed to software development. P.G. contributed to writing—review and editing, project administration, and funding acquisition. E.K. provided supervision, project administration, and contributed to funding acquisition.

Data Availability

Code and the sample dataset are publicly available on the Code Ocean platform (https://doi.org/10.24433/CO.5264887.v1). Here, it is possible to run the code directly and check its functionality on the sample dataset inserted by us. Other data-sets are available from the authors.