Preview Site Description

This page explains how to view available client sites and preview data description on each site to define a job

Each project undergoes through a different data description process depending upon its applications. These descriptions are useful to improve model performance in a federated learning scenario. You can modify your dataloader of each site in job definition using these statistics.

Following is a list of sample data description values we provide for chest x-ray data at each client site:

1. Data size

2. Class labels - Unique labels present, Histogram

3. Scan view (AP/PA/Lat) - Unique labels present, Histogram

4. Age group - Unique labels present, Histogram

5. Imaging hardware - Unique labels present, Histogram

6. Image quality (3 levels) - Histogram

7. Image type (DICOM/JPG/PNG) - Histogram

8. Locality (Pincode/city) - Unique labels present, Histogram

If there is any other particular description or assosciated statistics you need, please get in touch with us!

Example - NIH Chest X-Rays dataset

As an example dataset for Proof-of-Concept, we have shown data description for the NIH chest x-ray 14 dataset under Multilabel chest xray classification project on the platform. Under Public Site Description, you can see three sites - Site 1, Site 2 and Site 3. All three sites have subsets of the NIH data set. Data description for each site shows:

1. Total data size

2. Number of disease classes

3. Healthy:Unhealthy ratio

4. Histogram of class labels