Effect of an exercise program aimed at minimizing the compensatory movements, at the proximal level, of the affected upper extremity of children with Unilateral Cerebral Palsy during the execution of bimanual activities, evaluated with Artificial Intelligence. Randomized Clinical Trial

Unilateral spastic cerebral palsy (CPUE) produces an alteration of the contralateral hemicosis, with the upper limb (ES) being more affected and presenting motor deficits and postural control that make activities of daily life difficult. This last deficit originates from difficulties in bimanual performance and triggers bodily compensations during development. Scientific evidence indicates that work in the proximal part of this ES most affects the bimanual functionality of the child.

Executive Summary
There are 3 different phases in this project. First, the DeepLabCut (IA) tool will be validated and kinematic features will be extracted. Next, an expert consensus will be reached on the guidelines for exercises for the trunk and the proximal part of the EEHS. And finally, a randomized clinical trial will be conducted to test whether these exercises decrease compensations and reflect an improvement in bimanual function in children with PCUE.

Context and Justification
The project proposes a new methodology to objectively identify the compensations of patients with PCUE. The program agreed upon by experts and the subsequent verification of its correct functioning at home would represent a breakthrough for pediatric physiotherapy since it would lead to the introduction of a new focus of attention in the neurorehabilitation process with children with Unilateral Cerebral Palsy.

General Objective
Validate through deep learning (AI) the marker-free detection of movement trajectories in a remote program of motor exercises focused on the proximal part of the EESS of children with PCUE aged 8-16 years (levels I-II according to the "Manual Ability Classification System”) and from this information infer whether there is a decrease in compensations and improvement in bimanual performance.

Specific Objectives

• To validate whether the computational tool DeepLabCut (IA) allows correctly extracting scapular movement trajectories in patients with PCUE.
• To design the numerical and objective characteristics that allow to differentiate between patients with PCUE and those without neurodevelopmental pathology based on the kinematic record of the trajectories of the movement of the trunk and shoulder girdle in the pediatric population.
• Design an index based on kinematic trajectories to be able to objectively evaluate the compensation mechanisms in children with PCUE
• Establish an expert consensus on the exercise regimen designed to decrease compensatory movements of the trunk and the proximal part of the upper limbs in children with PCUE.

PHASE I

For data registration, 2 videos will be obtained from both subjects in which the trajectories of the movement markers will be extracted.

Features of video recording:

• Chamber model of the project
• Resolution of the chamber
• Camera position: Placed on a tripod 30 cm away from behind where the patient is sitting.
• Frame and center the video in the area of ​​both shoulders to the base of the glutes
• White and smooth background
• Position of the subject: Sitting on a stool without support and the feet must touch flat on the floor.
• The back plan has been chosen since it is where the compensations can be better observed.

The bimanual activities of daily living (AVD) to be developed will be the following:

• Unscrew the cap of a bottle.
• Put stick glue on a piece of paper.
• Zip up a jacket.
• Take off this jacket.
• Bring a tray.
• Comb the back of your head.
• Touch the lumbar part of the back.

The first 5 are part of the Childrens Hand-use Experience Questionnaire 2.0 test (CHEQ 2.0). These are activities of daily living that accentuate the decompensations in addition to covering all the movements that the scapula performs.

Inter-rater reliability will be assessed using DeepLabCut software, which uses semi-automatic labeling with a convolutional neural network (CNN).

The training of the neural network will be carried out by researcher 1 and 2, based on the manual labeling of n = 20 significant frames, obtained from a k-means clustering algorithm.

The intra-rater reliability will be assessed in the same way after the passage of one week.

PHASE III

A single-blind randomized clinical trial will be conducted.

For the bimanual performance variable, parents will be asked, together with their children, to fill out the CHEQ 2.0 form (https://www.cheq.se).

The results sheet will be downloaded and sent to the project email with the encrypted document (a video tutorial on how this process is carried out will be provided). If they prefer, they can physically bring it to the center.

For the analysis of the kinematic parameters, parents will be asked to film their child carrying out the activities to be evaluated, under the same registration conditions as in Phase I. They will have to send the file via SwissTranfer ( https://www.swisstransfer.com/es-es ) to the IP email. This will encode the file according to the randomization.

In the intervention of both groups, their usual treatment will be maintained. This is usually customized to the needs of this one at the time they meet. They tend to follow the same pattern: 1 session of physiotherapy and 1 session of occupational therapy per week, or combined, and they also usually have a follow-up with psychology. In addition, they all usually do between 1 and 2 extracurricular physical activities a week, in addition to school physical education subjects.

The intervention of the GE will consist of the addition to his usual treatment, the exercise program validated in Phase II. These exercises will be described on the Physiotec platform ( https://physiotec.ca/esp/es/ ). The subjects or their parents will have to mark the boxes as they are done.