Co-registering DAPI images is a critical step in the multiplexing workflow, particularly after tissues have undergone stripping and re-labeling. While visual inspection suffices for projects with manageable image sizes and sample numbers, it becomes impractical for quality control (QC) when dealing with a large volume of pixels. Therefore, an automated solution is required. This blog post will explore the procedure developed to automate the QC process for co-registering over 900 DAPI images from TMA cores.

Assessing the Alignment Between Images

To accurately assess the alignment between the fixed and moving images, we conducted calculations of the Similarity Metric and Pearson’s Colocalization Coefficient for all 900 image pairs. These metrics quantify the extent of overlap and colocalization between the two images, providing valuable insights into the quality of the co-registration process. The Similarity Metric, specifically using Mattes Mutual Information, is calculated iteratively during the registration process. As the algorithm approaches a solution, this value progressively decreases until the maximum number of iterations is reached or converges to a nearly constant value. A Similarity Metric less than -1 indicates a good alignment between the images. The Pearson’s Coefficient reflects the pixel overlap between the fixed and moved images, accounting for differences in overall intensity. A Pearson’s value close to 1 signifies a strong colocalization between the images. By plotting these two metrics against each other, we not only established a failure threshold but also gained valuable observations regarding potential imaging irregularities.

Successes and Failures

Let’s first study the most significant success and failure. Figure 1 displays the core that achieved the highest success, with a Similarity Metric of -1.7 and a Pearson’s Coefficient of 0.91. Before applying the transformation, a noticeable offset exists between the fixed (red) and moving (green) images. However, after the moving image is transformed, the composite image exhibits a bright yellow color, indicating successful overlap between the green and red signals. On the other hand, the core that experienced the most significant failure (Figure 3B), returned a Similarity Metric of -0.01 and a Pearson’s Coefficient of -0.04, due to inconsistent region selection during image acquisition..

After plotting the Similarity Metric against the Pearson’s Colocalization Coefficient for all pairs (as shown in Figure 2), a clear trend emerges between the two extremes, identified as core A and core B. Now, let’s shift our attention to core C, which lies between these two extremes. Despite having a satisfactory Similarity Metric close to -1, the Pearson’s value falls below 0.4. Despite this, the images for both panels in this core (Figure 3C) would still be deemed acceptable for image analysis. The decrease in the Pearson’s value can be attributed to missing sections within the core and the presence of an illumination artifact in the top-left region of the fixed image. This evidence suggests that cores positioned to the left of this data point also meet the criteria for acceptability. As we delve further into the investigation of cores positioned to the right of core C, it becomes increasingly evident that a distinct boundary can be drawn between the two clusters. These cores of interest are alphabetically labeled in Figure 2 and are classified as either success (green) or failure (red). Representative images of these cores can be observed in Figure 3.

By quite literally drawing a line between the acceptable and failed cores and extrapolating its function, the quality control (QC) constraints have been effectively established (Figure 4). Overall, only 5.3% of all registrations encountered failure, primarily due to tissue tearing, warping, or sample detection issues during the acquisition process. Importantly, registrations were still considered successful even when small sections of the cores were missing, or illumination artifacts were present. These metrics offer significant potential for further optimization of the co-registration algorithm, enabling expedited quality control procedures.

Visikol is home to a team of creative experts who possess extensive expertise in developing innovative and tailored solutions for large-scale data projects. If you want to discuss your current work and explore collaboration opportunities, we invite you to contact us. Our experts are eager to assist you in achieving your objectives.