Hello! I'm new to using QuPath and have been working on a project analyzing the expression of the aging marker p53 in the heart tissue of guinea pigs. The sample images are stained with DAB, which produces different shades of brown for p53-positive areas, while the negative tissue (Hematoxilyn) appears in shades of gray or blue, and the background is white. I am trying to quantify the percentage of the positive area for p53 within the tissue, excluding the background. I followed the tutorial from this https://youtu.be/kGvZRBEeqI0?feature=shared to get started.
Analysis Goals:
I want to measure the of positive % area (brown regions indicating p53) relative to the total tissue area in each sample (excluding white background areas) and accurately distinguish the positive % p53 areas from the rest of the tissue.
Challenges:
There are two main challenges:
Excluding the Background: I need to calculate the percentage of p53-positive (brown) areas in the tissue alone**, excluding any white background. (I think) this requires the program to recognize different types of browns for p53, different whites for the background, and various shades of gray or blue for negative tissue.
Color Variability Across Samples: I have 198 samples in total, representing different guinea pigs, tissue types (myocardium, endocardium, pericardium), and both ventricles. For each tissue type, I have 3 samples per guinea pig, which introduces even more color variability. To address this, I plan to create "representative canvases" for each type of tissue. For example, I’ll create one canvas with the 11 most representative samples of myocardium from the right ventricle across all guinea pigs, and another canvas with the 11 most representative myocardium samples from the left ventricle. I will apply the same approach for the pericardium and endocardium. This should help QuPath learn the color differences and apply them across the entire dataset, but it will take a lot of time training pixel classifier ...
Questions:
Does anyone have any suggestions on how to tackle the first challenge of excluding the background effectively? Also, do you think my proposed solution for the second challenge (using representative canvases) is a good approach for optimizing workflow time and reducing error percentage?
I am attaching images in dropbox of the tissue samples to help clarify my challenges. If anyone could guide me on how to proceed with these challenges, I would greatly appreciate it! I don’t have much experience with coding, but if it’s necessary to solve these issues, please indicate what’s required, and I’ll do my best.
Background:
Hello! I'm new to using QuPath and have been working on a project analyzing the expression of the aging marker p53 in the heart tissue of guinea pigs. The sample images are stained with DAB, which produces different shades of brown for p53-positive areas, while the negative tissue (Hematoxilyn) appears in shades of gray or blue, and the background is white. I am trying to quantify the percentage of the positive area for p53 within the tissue, excluding the background. I followed the tutorial from this https://youtu.be/kGvZRBEeqI0?feature=shared to get started.
Analysis Goals:
I want to measure the of positive % area (brown regions indicating p53) relative to the total tissue area in each sample (excluding white background areas) and accurately distinguish the positive % p53 areas from the rest of the tissue.
Challenges:
There are two main challenges:
Excluding the Background: I need to calculate the percentage of p53-positive (brown) areas in the tissue alone**, excluding any white background. (I think) this requires the program to recognize different types of browns for p53, different whites for the background, and various shades of gray or blue for negative tissue.
Color Variability Across Samples: I have 198 samples in total, representing different guinea pigs, tissue types (myocardium, endocardium, pericardium), and both ventricles. For each tissue type, I have 3 samples per guinea pig, which introduces even more color variability. To address this, I plan to create "representative canvases" for each type of tissue. For example, I’ll create one canvas with the 11 most representative samples of myocardium from the right ventricle across all guinea pigs, and another canvas with the 11 most representative myocardium samples from the left ventricle. I will apply the same approach for the pericardium and endocardium. This should help QuPath learn the color differences and apply them across the entire dataset, but it will take a lot of time training pixel classifier ...
Questions:
Does anyone have any suggestions on how to tackle the first challenge of excluding the background effectively? Also, do you think my proposed solution for the second challenge (using representative canvases) is a good approach for optimizing workflow time and reducing error percentage?
I am attaching images in dropbox of the tissue samples to help clarify my challenges. If anyone could guide me on how to proceed with these challenges, I would greatly appreciate it! I don’t have much experience with coding, but if it’s necessary to solve these issues, please indicate what’s required, and I’ll do my best.
LINK DROPBOX: https://docsend.com/view/s/ep3ycd6xpsszv76g
P.S. This entire message was translated using ChatGPT because English is not my first language lol