Added a new DICOMWebWSI class to allow access to protected DICOMweb stores.

[Steven-N-Hart]

Motivation and Context

Currently, the TRIDENT library excels at processing whole-slide images from local file formats (like SVS, TIFF) but lacks native support for cloud-hosted DICOMs. As modern digital pathology workflows increasingly rely on cloud infrastructure and standardized formats, this limits TRIDENT's direct applicability in these environments. Accessing slides via DICOMweb is a critical feature for scalability and interoperability.

This pull request introduces the DICOMWebWSI class, a new WSI object that enables TRIDENT to directly read and process whole-slide images from any DICOMweb-compliant server, such as the Google Cloud Healthcare API or Orthanc. This eliminates the need for manual downloads, allowing for seamless integration with cloud-based pathology archives.

Description

This PR adds the following key components:

1. DICOMWebWSI Class (trident/wsi_objects/DICOMWebWSI.py): A new WSI reader that inherits from the base WSI class. It connects to a DICOMweb server via a study/series URL and handles the complexities of the DICOM format for whole-slide imaging.

2. WSIFactory Update (trident/wsi_objects/WSIFactory.py): The factory has been updated to recognize http:// or https:// URLs and automatically instantiate the DICOMWebWSI class.

Key Features of DICOMWebWSI:

• Metadata Parsing: Automatically fetches and parses metadata for all instances in a DICOM series to determine dimensions, tile sizes, and pyramid levels.

• Pyramid Construction: Organizes DICOM instances into a logical pyramid, sorting them from highest to lowest resolution.

• *Flexible Frame Handling: Supports both tiled (multi-frame) and non-tiled (single-frame) DICOM instances, which is crucial for handling vendor variability in how lower-resolution pyramid levels are stored.

• Authentication: Includes logic to handle authentication for Google Cloud Healthcare API, with the ability to use gcloud credentials.

• Efficient Tiling: Implements a read_region method that calculates the correct frame numbers to fetch, composites the tiles into the requested region, and includes an in-memory cache for fetched frames to improve performance.

How Has This Been Tested?

The DICOMWebWSI class was tested extensively using the run_single_slide.py script against a whole-slide image hosted on a Google Cloud Healthcare API DICOMweb server.

[guillaumejaume]

Hi @Steven-N-Hart, many thanks for this PR.

DICOM web is quite specific. Some of the files/changes that you included would be best in a dedicated tutorial that people can check if they have a similar use case.

for instance,

• .env.example should be set per user (ok if in a dedicated tuto example)
• all additional dependencies should be optional for dicom web
• the code:

from dotenv import load_dotenv
load_dotenv()

should be set at inference time for users that need it.

Let me know if you have time to make these changes, thanks

[Steven-N-Hart]

Good idea. I can start on that next week. FWIW, I recommend adopting dotenv for the entire library because of your own dependencies on HF_TOKEN. Adding it as a full fledged requirement would save the user from an unexpected crash later in the code if the value is not set when first invoked.

…

________________________________ From: Guillaume Jaume ***@***.***> Sent: Sunday, October 19, 2025 9:47:51 AM To: mahmoodlab/TRIDENT ***@***.***> Cc: Hart, Steven N., Ph.D. ***@***.***>; Mention ***@***.***> Subject: [EXTERNAL] Re: [mahmoodlab/TRIDENT] Added a new DICOMWebWSI class to allow access to protected DICOMweb stores. (PR #164) External Email Notice: This message was sent by someone outside Mayo Clinic. [https://avatars.githubusercontent.com/u/8115947?s=20&v=4]guillaumejaume left a comment (mahmoodlab/TRIDENT#164)<#164 (comment)> Hi @Steven-N-Hart<https://github.com/Steven-N-Hart>, many thanks for this PR. DICOM web is quite specific. Some of the files/changes that you included would be best in a dedicated tutorial that people can check if they have a similar use case. for instance, * .env.example should be set per user (ok if in a dedicated tuto example) * all additional dependencies should be optional for dicom web * the code: from dotenv import load_dotenv load_dotenv() should be set at inference time for users that need it. Let me know if you have time to make these changes, thanks — Reply to this email directly, view it on GitHub<#164 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ACBKURJTSKNBJ5WXJKC7N233YOQBPAVCNFSM6AAAAACI3FWG7SVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTIMJZG4ZDGMJTGQ>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[Steven-N-Hart]

Tutorial added.

[Steven-N-Hart]

Any other changes?