Checklist NTO AI 2024 Stage 1 (qualifying), engineering tour · II.3.1 task
PN-expert: Parkinson's Exercise Classification
Russian title: PN-expert
Classify which motor-assessment exercise a Parkinson's disease patient performed, from recorded test data and patient metadata.
The task
Diagnosing Parkinson's disease relies on the MDS-UPDRS scale and on the physician's subjective perception. The PN-expert platform analyses videos of patients performing special exercises; to measure tremor correctly it must also recognise automatically which exercise is being performed. Participants build a model that classifies the exercise from observations of patients testing PN-expert.
For each sample the data give the name of the file with the patient's full test record, the target "Folder Path" (exercise class), sex, age in full years, the neurostimulator state (off/on) and the physician's MDS-UPDRS rating (0–5); the full records are provided in an archive.
Abridged and translated by SOTA from the official Russian materials. The official statement has the exact rules, and it wins wherever this summary differs.
In English
This task was published in Russian. SOTA translated its 2 files into English on 16 September 2026. Only the words changed in the notebooks: markdown, code comments, messages and printed output. The code, file names and paths are the original's, so a translated notebook runs with the original data.
- Task statement Russian original of Task statement
- Solution notebook Russian original of Solution notebook
Read the task statement in English
PN-expert: Parkinson's Exercise Classification
English translation by SOTA – AI Community of the Russian original. Organisers who would like this translation removed can email [email protected].
Source: the 2023/24 task collection of the National Technology Olympiad (NTO), "Artificial Intelligence" profile, Stage 1, pp. 85–86: original PDF.
Engineering tour
The task aims to test the participants' skills in working with classification algorithms in order to develop a model capable of automatically classifying the exercises performed by patients in video recordings. It tests the participants' competences in data processing and in applying machine learning algorithms, as well as their ability to immerse themselves in a subject domain and to apply formulas and classical algorithms when solving artificial intelligence problems. The task helps to assess participants on their knowledge and skills in programming, medicine, tabular data processing and mathematical statistics, and in applying technical solutions to real-world problems.
Task II.3.1. PN-expert (100 points)
Topics: programming, data exploration, machine learning, mathematics.
Problem statement
In the first qualifying stage, participants are asked to solve a simple task: determining the name of the exercise used to assess postural and kinetic tremor in patients diagnosed with Parkinson's disease.
Why does this matter?
Diagnosing and establishing Parkinson's disease (hereinafter PD) requires the physician to have sufficient experience and an understanding of the requirements of the clinical guidelines approved by the Russian Ministry of Health. The most widely used diagnostic instrument for PD is the unified rating scale MDS UPDRS: https://sudact.ru/law/klinicheskie-rekomendatsii-bolezn-parkinsona-vtorichnyi-parkinsonizm-i/prilozhenie-g1-gn/prilozhenie-g1/.
Consequently, establishing a clinically reliable diagnosis is influenced by a subjective factor that rests solely on the physician's perception of the patient's movements during the examination. In other words, this assessment is at present overwhelmingly subjective!
An objective assessment of motor disorders is a priority for the clinical follow-up of patients when establishing a diagnosis and when prescribing and adjusting treatment; it matters most when monitoring and controlling the patient's responses to changes in drug therapy (a change of dosage and/or of medication). Researchers developing PN-expert are working on such an objective instrument: https://inscience.news/ru/article/nti/8982.
This platform makes it possible to track the condition of patients by analysing video recordings in which people perform special exercises.
For the diagnostic instrument to work correctly, it is important that the algorithms not only measure tremor successfully but also automatically distinguish between the exercises that the patient performs in the video. In this task, we invite you to work with a large body of data from observations of patients testing PN-expert and to build a machine learning model that helps to classify the exercises.
Link to the original competition: https://cups.online/ru/contests/first_round_nto_ai.
Data description
Translator's note: the column headers in the data are in Russian; each is kept below as printed in the statement, followed by its English meaning in brackets.
- Имя файла данных (data file name): the name of the file that holds the complete records of the test results of the given patient.
- Folder Path: our target variable. The class of the exercise that the given patient performed.
- Пол (sex): the sex of the given patient.
- Полных лет (age in full years): the age of the given patient in completed years.
- Пациент off/on (patient off/on): the operating state of the given patient's neurostimulator.
- Определенный диагноз врачом (0–5) (diagnosis assigned by the physician, 0–5): the physician's subjective assessment of the given patient's condition on the
MDS UPDRSscale.
The archive with the complete records can be downloaded from this link: https://disk.yandex.ru/d/3sQ-vEewhCh2aw/data.zip.
[Figure: diagram of the 21 hand landmarks, numbered 0–20 and joined by lines; see the original statement, page 86. Its legend lists the landmarks as follows.]
| No. | Landmark | No. | Landmark |
|---|---|---|---|
| 0 | WRIST | 11 | MIDDLE_FINGER_DIP |
| 1 | THUMB_CMC | 12 | MIDDLE_FINGER_TIP |
| 2 | THUMB_MCP | 13 | RING_FINGER_MCP |
| 3 | THUMB_IP | 14 | RING_FINGER_PIP |
| 4 | THUMB_TIP | 15 | RING_FINGER_DIP |
| 5 | INDEX_FINGER_MCP | 16 | RING_FINGER_TIP |
| 6 | INDEX_FINGER_PIP | 17 | PINKY_MCP |
| 7 | INDEX_FINGER_DIP | 18 | PINKY_PIP |
| 8 | INDEX_FINGER_TIP | 19 | PINKY_DIP |
| 9 | MIDDLE_FINGER_MCP | 20 | PINKY_TIP |
| 10 | MIDDLE_FINGER_PIP |
Assessment criteria
Accuracy on the private part of the dataset.
Link to the description of the metric code: https://disk.yandex.ru/d/3sQ-vEewhCh2aw/scoring.ipynb.
Solution
You are required to develop an algorithm that classifies the type of movement and to present its output as a CSV file with the columns path and pred.
Answer: the solution to the task is a file with predictions, which is uploaded to the competition platform to assess the quality of the solution. An example of a solution file for upload: https://disk.yandex.ru/d/3sQ-vEewhCh2aw/sample_submit.csv.
Link to the expert reference solution: https://disk.yandex.ru/d/3sQ-vEewhCh2aw/reference_solution.
Translated by SOTA. The Russian original is the official version and wins wherever the two differ. The statement is task II.3.1 on pages 85–86 of the NTO 2023/24 task collection for the Artificial Intelligence profile; the solution notebook is the organisers' reference solution (reference_solution/solution.ipynb on Yandex Disk). Column headers and class labels in the data are in Russian; the statement and the notebook gloss them, and four large animated plots are left in the original notebook only. If you organise this olympiad and would like the translation removed, email [email protected] and we will take it down.
At a glance
- You get
train.csv/test.csvmetadata,data.zipwith the full records,sample_submit.csv, scoring code.- You submit
- CSV with columns path and pred, uploaded to the competition platform.
- Scoring
- Accuracy on the private part of the test set.
- Rules
- Individual task.
- Format
- Stage 1, engineering tour (100 points).