IFMBE Scientific Challenge Competition

Introduction

TITLE: “Segmentation Challenge: Wounds and Markers in Focus”

Welcome to the IFMBE Scientific Challenge Competition of the Joint 20^th Nordic-Baltic Conference on Biomedical Engineering and 24^th Polish Conference on Biocybernetics and Biomedical Engineering (NBC 2025 & PCBBE 2025). The participants of the current Scientific Challenge are asked to show a model for image segmentation able to distinguish wounds and linear scale markers from the background.

How to participate

• Register through the email sent to the following address: nbc2025_challenge@ibib.waw.pl providing the following information:
  • Name of the team,
  • Given names, surnames, titles /degrees and affiliations of all the team members (in the case of student members – the supervisor data should be also provided),
  • Name and email of the corresponding team member,
  • The statement: “All members of the team [NAME OF THE TEAM] consent to the processing of their personal data for the purpose of organizing the IFMBE Scientific Challenge Competition at the 20th Nordic-Baltic Conference on Biomedical Engineering and the 24th Polish Conference on Biocybernetics and Biomedical Engineering (NBC 2025 & PCBBE 2025)”
  • The email of the corresponding team member and addresses of students’ supervisors must be in domains of organizations listed as their affiliations, e.g. universities, research institutes, etc. (not in services such as gmail, yahoo. etc.). In the case your team is positively verified as eligible for the Challenge, you will get an email with a link to the training and validation datasets.
• Download the datasets and develop your model in Python for segmenting the images using ground truth masks. You may use a custom created model for the segmentation or pretrained models.
• Submit your trained solution in a self-contained executable file that we can run. The solution should look for the test dataset in a particular folder and generate the masks for each test file (see detailed description in “Evaluation of the solution on the test data set” section below). We will calculate your overall F1 scoring using our software.
• Submit an overview paper presenting your solution to the NBC 2025 & PCBBE 2025 through the webpage of the Conference at https://nbc2025.ibib.waw.pl. You will be notified if your paper has been accepted.
• Attend the Conference (NBC 2025 & PCBBE 2025, Warsaw, Poland, June 16-18, 2025), and present your work there during the IFMBE Scientific Challenge Competition session. Please note that at least one team member must be registered and attend the Conference.

The registered user will be responsible for not disclosing the downloaded datasets to third parties.

Please do not submit your solution to other conferences or journals before the conference ends. We are planning to prepare a review article with your best solutions to a prestigious journal co-authored by challenge participants and discussing their approaches.

The datasets

Train: 371 files, format JPG, size 512×512 pixels with corresponding masks as PNG files

Validate: 80 files, format JPG, size 512×512 pixels with corresponding masks as PNG files

Please cite the article:

“Foltynski P and Ladyzynski P. Internet service for wound area measurement using digital planimetry with adaptive calibration and image segmentation with deep convolutional neural networks. Biocybern Biomed Eng. 2023;43:17-29. https://doi.org/10.1016/j.bbe.2022.11.004”

in whichthese images were used for the first time in the AutoPlanimator Project (https://autoplanimator.ibib.waw.pl).

Structure of the folders

The training and validating datasets are in the following folders:

datasets\

    train\

               images\image001.jpg ………… image371.jpg

               masks\image001.png …………. image371.png

    val\

               images\image001.jpg ………… image080.jpg

               masks\image001.png …………. Image080.png

Example of the image and its corresponding mask are presented at Figure 1.

( A )

( B )

Fig. 1. Example of image with wound and scale markers (A) and corresponding ground truth mask (B).

The ground truth mask is composed only with 3 colors: black, i.e. RGB = (0, 0, 0), for wounds, gray, i.e. RGB = (127, 127, 127) for background, and white, i.e. RGB = (255, 255, 255) for scale markers.

The position of wound and scale marker on the predicted mask must be also encoded with the same colors as used for the ground truth mask.

Some images may present more than one wound and more than one scale marker. There are images without scale marker, but each test file will have at least one scale marker and one wound.

Evaluation

The predicted mask for each test file will be compared with the ground truth mask. We will use the F1 score for evaluating the model performance. First, an averaged F1_i score will be calculated for each predicted mask. Then a mean F1 score will be calculated for all test images. A team with highest mean F1 score will win the Challenge. Each team is allowed for three attempts.

Details of calculations

For the ground truth mask and the predicted we will calculate the number of true positive (TP), false negative (FN), and false positive (FP) pixels separately for all wounds at the image and for all scale markers. Then we will calculate the wound F1_w and the scale marker F1_s.

Fig. 2. Exemplary ground truth and predicted masks.

Fig.3. Combined ground truth and predicted masks presented in Fig. 2. Green color denotes TP pixels for the wound and for the scale marker; blue – FP pixels for the wound; black – FN pixels for the wound; white – FN pixels for the scale marker; yellow – FP pixels for the scale marker.

Taking into account colors at the ground truth and predicted masks we can define the TP, FN, and FP pixels by colors from the these masks. All definition are presented in Table 1.

Table 1. Definitions of TP, FN, and FP pixels by the colors at the ground truth and predicted masks.

F1_i for each test image will be calculated as mean of F1_w and F1_s , i.e.:

where F1_w is

and F1_s is

Finally mean F1 score for all n test files will be calculated:

Mean F1 score will enable to make a ranking of attending teams.

Test dataset

Test: 98 files, format JPG, size 512×512 pixels with corresponding masks as PNG files.

Evaluation of the solution on the test data set

The provided solution must be ready to make predictions, i.e., generate masks for the test files.

The test files will be stored in the folder “d:\temp\NBC2025_Challenge\test_files” as JPGs. The challenge software must generate predicted masks for all test files and store them in the folder “d:\temp\NBC2025_Challenge\predicted_masks”. The predicted mask should be a PNG file with the same core name as the test file. Example: for the test file “test001.jpg” the predicted mask file should be “test001.png”.

Important Dates

Personal data administrator and data processing information

The registration of a team for the IFMBE Scientific Challenge Competition is tantamount to expressing consent to the processing of personal data of the team members in accordance with Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free data movement and repeal of the Directive 95/46/EC.

The data is collected and processed by the M. Nalecz Institute of Biocybernetics and Biomedical Engineering PAS with its registered office in 02-109 Warsaw, ks. Trojdena 4 street for the purpose of organizing the IFMBE Scientific Challenge Competition of the conferences: the Joint 20th Nordic-Baltic Conference on Biomedical Engineering and the 24th Polish Conference on Biocybernetics and Biomedical Engineering (NBC 2025 & PCBBE 2025).

In all matters related to the processing of personal data, you can contact the Data Protection Inspector by e-mail iod@ibib.waw.pl.

All contact details and regulations can be found at: https://www.ibib.waw.pl/en/.