rmnldwg / lyprox Goto Github PK

It would be helpful if one could select multiple (or even just all) patients and then delete them all together.

If we want to include more datasets from different institutions and hospitals, I will need to include a model to support this.

So, create a model for an institution that can hold multiple patients (ForeignKey on patients, I think).

Recently, Django 4.0 was released and as far as I have understood the release notes, nothing major should break when I upgrade LyProX.

Especially on the dashboard view, tooltips on most elements would help people understand the functionalities quickly. Hence, this would reduce the need for an extensive (and tiring to read) documentation. What's left that needs explaining could be collected in a modal that pops up when one opens the page for the first time.

add function to export current selection of patients to CSV table. This CSV should have the same format as the one for uploading.

Planning CT is sometimes done after a neck dissection. The user needs to be informed that this might skew the displayed results.

Add a way/button to immediately select all N0 patients

I just found out that fontawesome has a pip package that can be used with django. However, there seems to be an issue with fontawesome 5 and pip. But I should probably sooner or later upgrade to fontawsome 6 anyways, so when that day has come, I can probably switch to the pip version.

Those levels that are subdivided (LNL I & II), also display the diagnostic results for the superlevel. So far, only the sublevels a and b are reported.

Also display stage prefix c and p in the patient's TNM stage.

For the central patients, make the most involved side the ipsilateral one

Include the option to select patient gender in dashboard.

Implement logging on multiple levels, so that - upon deployment - one can track which kinds of bugs and issues appear

Now that the interface is running on a server, I guess there's no real need anymore for the detailed explanations of how to install it locally.

It surely makes sense to keep them somewhere, but the README might be better suited for a general overview of the purpose and implementation of the interface.

Bertrand had the idea to display how many patients of each institution are in selection. This is helpful and necessary, I think.

Implement an authentication system so that we can safely deploy this interface on the web without the need to remove parts

Make the displayed institution buttons dependent on whether or not the user is logged in. So that datasets of hidden institutions are only shown after a login.

e.g. when selecting "midline extension" (+) as the only filtering criteria, Bertrand's and this interface are off by a count of 2. Other cases must be checked as well

Add some sort of versioning (e.g. with versioneer) so that we and others can keep track of changes more easily

add a button to the dashboard that allows the user to toggle between absolute numbers and percentages

Problem

The CLB data does not specify the modality (MRI, PET, CT, FNA) that was used to diagnose an LNL as clinically involved or healthy. It is already a consensus formed by experts. Now I need to combine this with our USZ data where we have detailed information on every modality.

Idea

One idea is to create an "umbrella button" called diagnostic consensus or something like that. Internally, that would be handled just like any other modality. But in the interface, selecting this modality automatically includes MRI, PET, CT and FNA as well - which is basically the diagnostic consensus for patients like the ones in our dataset.

When three diagnoses are available for a given lymph node level - e.g. [True, False, None] - then python's built-in all (which I am using for the logical AND) returns correctly False. However, this is not quite what we want. If a diagnostic modality has nothing to report w.r.t. to that LNL, we would like to ignore it.

I think, the solution would be to use ~any(~[True, False, None]) (doesn't quite work like that, because one cannot negate an array), since from a logic standpoint, this is the same and it treats the None like we want it to be treated.

After clicking on the "Choose a file..." button and selecting a file from the explorer, the webpage does not display that file's name.

Split into:

1. oropharynx: base of tongue, tonsil, rest
2. hypopharynx: all
3. larynx: glottis, rest

When a CSV file is uploaded that has the wrong date format for one or more fields, the whole thing crashes and throws a 500 error.

Ideally, the interface should be able to handle a variety of common date formats and if it encounters an unknown one it shouldn't crash, but gracefully inform the user of it.

Use Bertrand's interface as a guideline for testing the correctness of the dashboard.

It is unnecessary to keep the information on which side (right or left) the tumor occured. So, it'd make sense to discard that information and transform the side of the diagnoses to ipsi & contra, which is the part of the information that actually matters.

Problem

The logical OR and AND are not ideal or informative in every case.

Solution

Implement one or two new methods that combine the available diagnostic modalities in a smarter way.

The core data of an added patient cannot be edited. Only their tumors and/or LNL involvements can be modified.

Due to the hash-value check in the PatientForm it is not possible to edit a patient and give him the same name & birthday as before. One should add another check to make sure this hash-value comparison is only done upon patient creation.

Specifically, when I select modalities PET & pCT and look for patient without involvement in LNL II but with involvement in LNL III contralaterally, I see patients that do have involvement in LNL II (namely patient 127). That's bad.

Problem

Right now both the database as well as the migrations of the server and my local machine are completely independent. This means that if I change a model field and come up with an elegant solution to migrate the old data to the new model, I need to get that migration over to the server manually and also manage when and how to pull commits.

Possible Solution

I add the migrations of every app to what git tracks and write a custom management command that creates backups of the database before every migration. Also, I should stop performing the python manage.py makemigrations command automatically when commits to main get pushed to the server.

That would allow me to keep the general anatomy of the remote and local database in sync without syncing the database itself (which would be a security issue, I think).

It seems as if the interface crashes when the database is empty.

Indicate to the user when the selection of patients and diagnoses deviates from what was computed when "Compute" was pressed the last time. Maybe grey out the button after the computation until anything else is pressed.

When one deletes the tumor of a patient, that patient's overall T-stage does not get updated. E.g. when the deleted tumor is T3, but the remaining tumor is only T2, then the patient will still display T3 as its T-stage.

Right now, one cannot look at the next or the previous patient when in the detail view. Also, it's not possible to display a subset of the dataset in a list (e.g. the one that was selected in the dashboard).

In TNM staging, N-category has more details like a, b, c. I should extract and include that in the display.

The way the number of unknown patients is displayed in the dashboard is a bit confusing. Maybe drop that number altogether or also pull the number for healthy and involved inside the bar too?

If none of the T-stage checkboxes is selected and one hits "compute", the app raises a KeyError

smoking status, HPV status & neck dissection should also show numbers like the involvement

Replace date of diagnosis with "date of histological diagnosis"

Make it somehow possible to switch seamlessly between development and production without interfering with git source control.

In the detail view of the patients, describe which TNM edition was used.

For further anonymization, one could shift all dates by a couple of days (ideally a random number for every patient). By "all dates" I mean diagnose date and modality dates, not the birthday however, since that will be deleted anyways and is used for the hash function.

On first glance, the dashboard is still quite overwhelming. An idea could be to add a visual guide that - upon the first visit or when one clicks on the help button - puts the focus of the user to the individual parts of the interface one by one, explaining them in some detail.

Ideally, I can do this completely without cookies...

It looks like the filter for neck dissection is not working. For PET and neck dissection + I get 0 patients

Add a django app that provides an interface for the risk computation. This would require the ability to upload a batch of sampled parameters or run the sampling for a stored dataset.

If a patient has had an FNA, the logical OR introduces a lot of "unknown" labels in the LNLs that were not inspected by the FNA. But "unknown" & "True" for example should always yield "True".

Sometimes we would like to look at a dataset that we do not want to show to the world yet (or are not yet allowed to), so an option to hide e.g. datasets from certain institutions would be helpful there.

Define a shortcut for the computation. E.g. ALT + C or something like that.