QuPath vs Halo vs Visiopharm

Hi all,

In my lab, some people want to buy an image analysis sofware to analyse brightfield and fluorescent multiplexes.
I objected that QuPath is free, but they fell that to pay means better/easier/more reliable software. They think that using halo or visiopharm is straigtforward and will ease our workflow.
I’ve never used Halo and Visiopharm.
Are they that easy to use (even for complex multiplexes)
Do some people here have tried one or both and could give me the pro and cons compared to QuPath ?

Thanks

Nico

1 Like

Clearly I’m not an unbiased observer, so I won’t say much :slight_smile:

But, if you haven’t already, you could share this comparative study with your colleagues: https://doi.org/10.1038/s41374-018-0123-7

However, I personally wouldn’t discourage anyone from choosing commercial software if that’s what they prefer. I would argue that software isn’t necessarily better/easier/more reliable just because it’s expensive (as shown in the comparative study)… but it may very well be better/easier/more reliable for some applications/people/groups.

The main advantages of open source aren’t that it has no financial cost, but rather that it is completely open – you can check/modify the code, reproduce results, share methods freely with others who might not be able to afford licenses. It’s a better choice for open and reproducible science. But it doesn’t necessarily give a better user experience, especially for someone who will never look at the code or want to modify/script anything.

In my opinion, the worst choice is in-house developed stuff that is neither open nor commercial… since then no one can reproduce the analysis, even if they would have the money to pay for licenses. I’d say commercial software is at least better than that.

(Of course, if more people put their money into funding open source software then it could be better/easier/more reliable for everyone *cough cough*…)

5 Likes

I can’t comment on HALO, but have experience with both Visiopharm and QuPath.

+Visiopharm in Europe has Oncotopix or some such which can be used in clinical studies, that could be a factor.
-It is very expensive.
±If you need deformable whole slide alignment, it is fairly good at that, though even with re-stains on the same tissue slice, it was not good enough for single cell overlap.
-Multiplex brightfield is very frustrating as it does not support brightfield for it’s phenomap implementation (multiplex analysis).
-You cannot export the aligned images or really export much more than a snapshot of the screen.
-Objects you create can be exported as MLD files which can be imported into… other copies of Visiopharm. (See additional information in post below)
+Data exports to CSV/Excel as you might expect.
-----I find the whole APP design process a bit frustrating as finding the % positive of cells of class “A” out of A, B and C involves something like 5 steps, none of which can be coded. Repeating the same processing steps for 4 different labels often requires creating the same process 4 times, rather than “this is the process, apply it to these 4 labels.”
++However, the cell detection options are incredibly complex and powerful, and it has several deep learning options that are far better than QuPath’s pixel classifier at detecting complex objects (glomeruli, etc). Slight -, the deep learning options are only pixel classifiers, and do not do semantic segmentation. The post processing can be used to approximate this, but you need to be careful about how you split objects, still.

If you get Visiopharm, try and make sure your entire workflow can be done within Visiopharm, or you can accept an output that is a CSV file with your results for your next steps.

I believe @mesencephalon might have tried out Visiopharm recently, and has experience with QuPath as well, and might provide further thoughts…

1 Like

Hello Nicolas, it was a pleasure to speak with you earlier this week.

Likewise, being under the employ of Visiopharm, I will leave commentary up to end-users, but I find myself largely agreeing with @petebankhead’s very balanced arguements. There is also the question of how much development time you have to commit to projects for you and others to consider.

Aside from comparing capabilities, one important difference I would highlight is in terms of support and training. I love the community at image.sc but if you need precise and immediate expert support on a project, you don’t always have access to that with community help. Same goes for training.

You have my contact email, I’m happy to discuss any of these points or capabilities in more detail.

@Research_Associate: just to comment on a couple of your points:

You can export all ROIs and labels as MLD, XML or BigTIF files.

Thank you for highlighting these. I consider the DL tools a significant strength of the Visiopharm platform.

3 Likes

Could you please tell us a little bit more about your multiplex assays? It would make better sense to talk on your needs.

Hi Guray,

We produce brightfield and fluorescent multiplexes. At the moment, We want to get phenotypes of cells and compute some spatial measurements (distance to a compartment, distance between markers…)

Nico

Fluorescent multiplex means multiplex IF using OPAL fluorophores and vectra scanner and brightfield ones are consecutive staining based multiplex IHC methodologies like MICSSS?

I am asking because some commercial softwares have easy tools for different things worth to check if you don’t have computational background for editing scripts etc.

Well, we use Opal fluo because we can diluate them, but because vectra scanner doesn’t allow whole slide analysis, we often use our Olympus VS120 scanner (although this implies we can do less markers and we can’t substract autofluorescence).
For BF, up to now we used “classical” multiplex because we couldn’t align markers consecutively stained.
The purpose is to process automatically large cohorts of slides.

Nico

Vectra3 should allow whole slide analysis, though you have to build the whole slide images from the fields of view. Which Pete very nicely wrote a script to do in QuPath automatically on his Gists site.

I believe HALO also has the capability to stitch Vectra fields of view into a whole slide image.

We did it also in my lab 4 years ago, but scanning the whole slide costs ~1TB of space (if you want to keep the raw data) ; a little too much for our servers :wink: + the inform analysis time wich is quite long…

Hmm, not sure about all of the data storage steps, but I think most of the final slides were around 40-60GB IIRC. I can understand the concerns, but that does mean it “can” do whole slide analyses… if you really wanted to. And some labs are using QuPath for just that.

Those types of multiplex files are also where Phenomap within Visiopharm can come in handy. It has the same sort of pipeline as multiplex analysis in QuPath, but some added features like better cell border detection using various channels and the ability to use X% of the brightest pixels per channel are quite nice, and are not currently built into (the GUI) of QuPath.

I tried Visiopharm for its ability of whole slide image registration, however it is a hit or miss and does not work perfectly. We needed a high-throughput perfectly working software doing that but it was not a good candidate for high-throughput due to some manual QC steps. You can build workflows in Visiopharm and use it on your project images. It has many features for nuclear segmentation and cell classifications in addition to deep learning modules and others, however it does not have the same flexibility of Qupath and ease of use. (I really like their nuclear segmentation modules). It would worth the money for specific aims, otherwise Qupath can handle most things and free (thanks to Pete).
For your specific aims, if you have large volume of trials that needs to be analyzed with ease without scripting, coding and designing a pipeline (without a special effort), I would go for a commercial software. Otherwise it is a joy to work with QuPath, spend time on things you can achieve and build from scratch.

2 Likes

Thanks to all for your opinion.
Any Halo user here to give pro and cons ?

Nico

Hi Nico,

I have tried all 3 programs. I try to sum up some key points.
I agree with Visiopharm users. It’s really powerful program but will require a super user to oversee/ create APP for users (but you will also get great support and training). I tested it for a few weeks and a huge plus is the phenomap; it worked great to cluster the dataset with 30 markers. It uses a very smart for separating a real signal from over-spill in a dense tissue.

QuPath gives lots of flexibility, especially with scripting possibilities. We use custom model from StarDist within QuPath for dense tissue. It’s very intuitive and has great tutorials, also forum/Pete and the team is amazing for help and advices (Thank you!) and many users get independent very quickly. I like object classification a lot. We are currently training classifiers for a bigger dataset and some ‘simple’ 4 markers studies. It’s the best program if you require flexibility and want to test many options, I am also going to test QuPath<->CytoMAP combination. And I agree with Guray, it’s lots of joy to work with QuPath.

We also have some projects with Halo. I see lots of similarities to QuPath. It’s also user friendly, segmentation work well. For 30+ markers it takes some time to set up thresholds for each marker (but that’s true for each software), but you can save your analysis and reuse for all samples within your experiment. The disadvantage is that it currently does not support unsupervised clustering/ discovery mode. You can set phenotypes based on your markers: saying Class1 = Marker1+ & Marker2+ & Marker3-. Once you have your supervised clusters, you can use their spatial analysis module with some distance analysis, proximity, density measurements.

it really depends on the specific project and your final question.

Please let me know if you have any other questions.

4 Likes