SAND

Why we need SAND?

This challenge stems from the need to analyse noninvasive, objective, and scalable biomarkers, such as speech signals, for early diagnosis and longitudinal monitoring of patients suffering from neurodegenerative diseases. This is because diseases such as Amyotrophic Lateral Sclerosis (ALS), present complex diagnostic challenges due to heterogeneous symptom profiles and overlapping clinical features.

Be part of the change!

With this challenge, we would like to redefine neurodegenerative disease assessment by positioning speech as a central, AI-powered biomarker for diagnosis and monitoring. We invite you to participate in the SAND challenge with your contribution. The five highest-ranked teams will be invited to showcase their work at IEEE ICASSP 2026. Accepted contributions will be published in the official IEEE ICASSP proceedings (IEEE-indexed). The challenge-dedicated session will highlight presentations from the top-performing participants and conclude with a panel discussion. Don’t miss this engaging event!

Ranking 1

Guidelines for participants

• Participants may submit up to three predictions per task, with only the final submission being evaluated.
• Submissions may be made to either one task or both. Each submission must include a short (2-page) description of the methodology used. At this stage, although not a mandatory requirement, it is strongly recommended that you draft the 2-page paper using the Word or LaTeX template provided in the IEEE ICASSP PaperKit 2026.
• Each participant may only be part of one team.
• Each team must register to participate in this challenge.
• During the registration, please, check carefully the entire list of Team Members, because no modifications will be allowed in the future. You will not be able to add additional members to the 2-page description paper, which will be published in the IEEE ICASSP proceedings if it is accepted.
• Each team can download the training dataset from the dashboard.
• At the appropriate time (see guidelines), each team can download the testing dataset from the dashboard. Use it to obtain the results to be submitted.
• Test data must not be used during training. All parameter tuning should be performed using the training set, from which validation sets may be created.
• Each team can submit its results from the dashboard.
• Submissions that are incomplete or contain missing entries will be excluded from evaluation.
• The final rankings will be published on this platform, and the winners will be notified.
• Rankings will be based on the best achieved performance for each individual task.
• To determine the winning method(s), the top 3 highest-performing submitted models (for each task) must provide an executable file (or a notebook) in order to reproduce the classification/prediction results. We strongly encourage all teams to release their code publicly on their profiles.
• The five top teams will be selected, taking into account the distribution of participants across tasks: the top 2 teams from Task 1 and the top 2 from Task 2 will be chosen, with the fifth slot going to the third-ranked team in the task with the highest number of submissions.
• The five highest-ranked teams will be invited to submit a 2-page paper to be presented at IEEE ICASSP 2026.
  The 2-page paper must be formatted using the Word or LaTeX template provided in the IEEE ICASSP PaperKit 2026 and submitted by the camera-ready deadline.
• Our baseline implementation includes predefined validation sets (20% of the training set), which may be used during training. Details of the composition of our baseline training and validation sets can be found in Dataset section
• The use of external resources (training data and/or pretrained models) is allowed if the data or pre-trained models are publicly and freely accessible, all resources are available before the challenge begins, and any dataset or model used is properly cited in the 2-page description of the methodology.
• Models developed using private data are not allowed.

where \(TP_c\) is the number of true positives, \(FP_c\) of False Positives, \(FN_c\) of False Negatives, all reported for class \(c\), while \(|C|\) is \(5\) for task 1, and \(4\) for task 2.

Task 1 - Multi-Class Classification

• ALS with Severe dysarthria (Class 1)
• ALS with Moderate dysarthria (Class 2)
• ALS with Mild dysarthria (Class 3)
• ALS with no dysarthria (Class 4)
• Healthy (Class 5)

• Class 1: 2.2%
• Class 2: 9.55%
• Class 3: 20.95%
• Class 4: 27.94%
• Class 5: 39.33%

Baseline

Task 2 - Prediction

• ALS with Severe dysarthria (Class 1)
• ALS with Moderate dysarthria (Class 2)
• ALS with Mild dysarthria (Class 3)
• ALS with no dysarthria (Class 4)

• Class 1: 13.64%
• Class 2: 21.97%
• Class 3: 28.79%
• Class 4: 35.60%

Baseline