Quantitative Analysis & AI

Digital slides are only as useful as the data you generate from them. With the emergence of so many new AI and image analysis platforms over the past three years, pathologists and scientists are spoilt for choice.  It can also be overwhelming if you are not 100% positive about what you need out of your image analysis and AI workflow. Here are a few questions you should consider in your evaluation.

• Stains, biomarkers, and modalities used Do you need image analysis on H&E, IHC, fluorescence, in-situ hybridisation? Which tissue types will be important? Do you require tumor detection or detection of specific morphologies? Can your proposed solution handle your range of applications?

• Cell or tissues? Do you need to investigate your images at the tissue morphology level, at the cellular level or both? Some image analysis providers offer a wide variety of tissue and cell-based applications in a single package providing different algorithms for the investigation of nuclei, membrane, cytoplasm as well as for detecting tissue texture and morphology. Other solutions may be specific only for a defined set of applications or tissue types.

• Application-specific analysis or toolbox solution? If you are in a clinical laboratory, most likely you will be looking for application-specific algorithms which are tailor made for a particular workflow (more on that later). While this works well for clinical applications, it may not be the best solution for research laboratories who need to address a wide variety of applications. In the latter case, it may be better to consider a “toolbox” type of solution where algorithms/AI can be modified to fit a larger number of tissue types and address a wide variety of research questions.

• Region of analysis or whole slide Do you expect to manually define a region of interest for your image analysis or conduct whole slide analysis? If so, you may want to check that your proposed solution offers analysis over multiple regions or the entire slide.

• File formats Digital slides exist in numerous file formats, some standard, some in common use and yet others specific to a single vendor. What will be the source of input files for your image analysis process? Do you want to be able to share these files with users on or off campus? Do you want to receive inputs from different scanning platforms?

• Multi-stage analysis Do you want to be able to ask a hierarchy of questions during your analysis. For example, would you like to be able to automatically detect an area of suspected tumor tissue and then conduct further cellular analysis within that specific region. Some solutions may not permit this type of multi-stage investigation.

• Batch analysis Does the proposed solution allow image analysis tasks to be queued for later processing? If not, you may need to wait some considerable time before conducting your next analysis.

• Ease-of-use How intuitive is the software interface and can you achieve your goals with minimal training. How much training does the vendor provide and what are the experiences of other customers? How easy will it be to bring other users on board and who will be responsible for the training process?

• Reliability Are the image analysis results reproducible? Can you run the same analysis multiple times and achieve the same results? Do you feel confident that the algorithms are robust and reliable? Are there reports of variability from other users?

• Users and accounts How many users will need access to the image analysis package? Does the proposed solution restrict the number of user accounts or simultaneous users? Will you need access from a single or from multiple geographic locations? Can results be freely shared with external collaborators?

• Local or cloud-based installation Will the analysis be performed through a local client-based package, through your own server or through a cloud server? Remember that server-side analysis will generally be more powerful and more costly, but will you need to move the image files to a different location after scanning? For remote analysis packages, what connection speed is required? Could there be any networking or latency issues?

• Time to perform analysis How fast does the analysis run and with what computing power? Does the platform allow you to scale-up your throughput when you are ready? Will you need to add additional local hardware, memory or a GPU to achieve your goals?

• Open-source solutions Are there open-source solutions which could solve the problem? If you have a single application or need an initial proof of concept, then an open-source solution may be suitable. Open-source solutions may allow you to test your theory before committing to a commercial package or may even be able to handle your entire project.

• IVD Algorithms. If you are in a clinical lab and looking for quantitative tools for clinical decision support, it is critical to make sure that the algorithms have been cleared, certified or registered for clinical use in your specific country/region. When it comes to quantitative tools for clinical use – read the fine print! Algorithms are generally only approved for a very specific (read limited) workflow – using a specific reagent kit, autostainer, slide scanning platform and even image management system/viewer. It is worth investigating other options. For example, CLIA laboratories may be able to self-validate some of their workflows, and thus have more flexibility in choosing solution provider(s).

• Customizing algorithms. Stained tissue can be very variable in intensity of stain, granularity, texture etc. Does your proposed image analysis solution use fixed algorithm parameters or can you customize those parameters according to your own slide quality and variation? Can you save customized algorithms for later use or to share with colleagues?

• Output data. In what format is the image analysis output? Does it include cell by cell data and statistical analysis tools? Can you export your data to other software packages for subsequent analysis?

Use the filters on this page to help you identify solutions that fit your requirements. You can hit the “compare” button to view specifications of up to five solutions side-by-side.

The specifications on this website are gathered from publicly available online resources; therefore, we cannot claim to be 100% accurate. It is always a good idea to make enquiries with the solution provider if you have doubts or unanswered questions.  If the solution provider has registered with Pathology News, then there will be a “send enquiry” button available right from the product page to make this even easier for you.

Good luck in your search!

Pathology News Team