AI Tools/Software To Analyze Biological Samples

Following the advice of @Cole Delyea, I consolidated and organized all the useful feedback on analyzing biological samples that I received in response to one of my questions in another post.

Background. During my PhD, one of the most challenging parts, has been analyzing several images for different purposes. Considering the huge impact of AI there are several tools that can be helpful to automatize these analyses, resolving issues of time consumption and human bias. In my post I shared some of the applications I believe these software can be really helpful.

Quantification of the number of vessels, including their diameters. This involves measuring the number and sizes of vessels in an image.
DAPI cell quantification. DAPI staining is a common method for visualizing nuclei in cell samples.
Particle quantification in confocal images and counting how many of them are inside the cells. This is particularly useful in studies involving the uptake of particles by cells, where it's essential to quantify not just the particles themselves but their distribution in relation to the cells.

Overview. Below is a list of some of the most useful AI tools and software, along with their key features.

1. QuPath

Type: Open-source
Features: Designed for whole slide image analysis, QuPath offers robust tools for cell counting, tissue segmentation, and stain quantification. It has an user-friendly interface and strong community support, which is invaluable for troubleshooting and sharing insights.
Best For: Tissue analysis and quantification, with a slight learning curve for utilizing advanced features through Python scripting.

2. ImageJ/FIJI

Type: Open-source
Features: A versatile platform that, while not AI-based by default, offers numerous plugins that employ machine learning for automated analysis. It's particularly effective for tasks like cell counting and particle analysis, with a wide range of tools for image preprocessing and analysis.
Best For: Researchers looking for a customizable analysis approach with extensive plugin options.

3. HALO

Type: Commercial
Features: Utilizes AI for quantitative pathology, capable of sophisticated analyses such as vessel detection and other complex tissue quantifications.
Best For: Advanced research projects requiring detailed and precise quantitative pathology analyses.

4. Aiforia

Type: Cloud-based (Commercial)
Features: Enables users to create their own deep learning models for image analysis without needing coding skills. It's a powerful solution for custom model development, offering scalability and accessibility through its cloud-based platform.
Best For: Researchers requiring customized deep learning models for image analysis without deep programming knowledge.

5. CellProfiler

Type: Open-source
Features: Designed for high-throughput image analysis, it can automatically count cells, identify fluorescently stained cell nuclei, and more. It stands out for its ability to manage large datasets efficiently.
Best For: Large-scale image analysis projects requiring automated cell counting and feature extraction.

6. Ilastik

Type: Open-source
Features: An interactive tool that's great for segmentation, classification, and tracking. It's user-friendly and requires minimal coding knowledge, making it accessible for a wide range of users.
Best For: Segmentation and classification tasks where user input and correction are beneficial.

7. Imaris

Type: Commercial
Features: Offers comprehensive functionalities for visualizing, segmenting, and interpreting 3D and 4D microscopy datasets. Its 'Spots' function is particularly useful for detecting and counting objects, such as cell nuclei.
Best For: Advanced 3D and 4D image analyses requiring detailed visualization and quantification.

8. Visiopharm

Type: Commercial
Features: Provides AI-driven image analysis for research and diagnostics, with capabilities for automated quantification and detailed tissue analysis.
Best For: Diagnostic and research applications requiring high precision and AI-driven analysis.

Extra 1. @John McSpedon showed a method for DAPI counting. He developed an internal dashboard. He recommends starting with the AGAR public dataset, for those interested in refining their DAPI counting approach. This dataset is valuable for fine-tuning Meta's Segment Anything model to enhance analysis precision. Furthermore, Bo Wang's lab has effectively demonstrated the potential of this model in medical image domains, achieving surprisingly high performance through transfer learning from natural images to medical applications, as detailed in their publication. This approach offers a novel perspective for those seeking advanced methodologies in DAPI counting.

Extra 2. @marco farina has successfully processed Hematoxylin and Eosin (H&E) images along with Lectin images to train a neural network algorithm for diagnosing chronic wounds. They followed a training and test dataset ratio of 85:15, in accordance with literature recommendations, and conducted validation with the provided image among others. The results were promising, showcasing the software's capability to automatically analyze the images. However, Marco mentioned the inability to plot the vessels' dimensions due to the absence of a scale bar and uncertain magnification, suggesting a possible 20x magnification but seeking confirmation. For further refinement and the addition of new functionalities, a larger dataset would be necessary. Marco expressed optimism about this being a positive beginning and invites feedback on the findings.