neonAtal motoR paTtErn autoMatIc analySis

Neonatal Care and Newborn Examination:

Every newborn undergoes a comprehensive clinical examination in the delivery room before being transferred to the postpartum ward with their mother (decree of October 18, 1994). At least one additional full examination is conducted during the maternity stay. Mandatory before the 8th day of life, this examination is recorded in the health booklet and allows for the issuance of the first health certificate (8th day certificate). It is performed in the presence of the mother or both parents, in a well-lit, adequately heated room, preferably before a meal or during a care routine, when the child is alert, and following hygiene rules (handwashing, use of a dedicated gown for the newborn, and disinfection of equipment). Significant clinical variations are possible among newborns, and a number of minor anomalies may be detected.

The neurological assessment of the newborn begins with gathering maximum information about prenatal history and birth context. Family and social backgrounds are also crucial. For instance, previous neonatal deaths in the family may indicate autosomal recessive metabolic disorders, while details of any previous maternal miscarriage are particularly relevant in suspected X-linked dominant disorders, where male lethality is common. Medical records are extremely useful sources for Apgar scores, blood gas results, antepartum monitoring data, and any obstetric complications. Clinical information regarding the immediate postpartum period, known as the "golden hour" in high-risk groups, can also be very valuable. Additionally, nurse observations, when available, are generally invaluable. When evaluating newborns with paroxysmal disorders, video sequences of the events in question can provide valuable diagnostic clues.

Clinical examination remains a central element and perhaps the most cost-effective and safest tool for evaluating newborns suspected of having a neurological problem ("Wusthoff CJ. How to use: the neonatal neurological examination. Arch Dis Child Educ Pract Ed. 2013;98:148-53." Extract from: Mustafa A. M. Salih. "Clinical Child Neurology." iBooks.). Thus, the main objectives of the examination are to determine whether the newborn is "neurologically normal" or not and to initiate further investigations if necessary. It also serves to assess the worsening (or improvement) of the newborn's condition over time. It should be noted that the newborn's neurological status often cannot be fully determined by a single neurological examination, especially if it is performed quite early. In fact, the predictive value of the neurological examination tends to improve with the newborn's chronological age, as illustrated by the case of a newborn who suffered significant hypoxic-ischemic injury at birth, for example. Indeed, this child's clinical presentation will vary over time, and a complete and accurate assessment of lesions is often possible remotely. Furthermore, it is well known that the newborn's neurological status can be affected by a range of "non-neurological" factors such as medications administered to the mother during pregnancy, infections, respiratory disorders, and unpleasant stimuli such as pain (from the cold hand of an examiner, for example) and hunger.

Description of the Research Focus There are several standardized and validated scales and approaches widely used in various neonatal units. Each has its strengths and weaknesses, but they are all comprehensive enough to provide an idea of the neurodevelopmental profile of both full-term and premature newborns. Many of them, such as the Dubowitz examination, are freely accessible on the internet and in most neonatology manuals.

Given the unique characteristics of the developing neonatal nervous system, it is necessary to adapt the formal neurological examination applicable to older children and adults. Observation is the first step of the examination, and it is crucial to spend as much time as possible observing the baby and their interaction with the environment ("Dubowitz L, Mercuri E, Dubowitz V. An optimality score for the neurologic examination of the term newborn. J Pediatr. 1998;133:406-16." Extract from: Mustafa A. M. Salih. "Clinical Child Neurology." iBooks.). This observation step is paramount in assessing the child and superior to the so-called manual examination, which involves palpation and directed manipulation of the child. This visual examination includes several registers related to the position and spontaneous movement of the child, both in terms of limbs, head movements, eye movements, and vocalizations. Interactivity is also crucial, as it indicates whether the child responds to voices and if automatic tracking mechanisms are in place.

This newborn observation step is crucial. It allows for the evaluation of cardinal functions such as motor skills, reactivity, and interaction, and provides insight into the functioning of the patient's central and peripheral neurological structures. It is important to understand that these elements allow for a much more comprehensive evaluation of the newborn, and that these same functions will be impaired during systemic pathophysiological phenomena for which neurological involvement is only secondary or indirect, such as sepsis, metabolic diseases, anoxic-ischemic encephalopathies, etc. This is why this clinical examination, even if very generic and not highly directed, is of crucial importance in the newborn's progression before discharge.

Best practices often appear to be dependent on the caregiver. Although the use of standardization in observation and data collection is unavoidable, clinical examination remains a phenomenon, an experience that includes a subjective element linked to the practitioner's experience and competence. Its expression thus exhibits a certain variability among patients. In addition to interindividual variability and variability in the evaluator's assessment, there is the difficulty of evaluation related to the child's age.

Therefore, the observational component of the clinical examination in neonatology relies on the combination of three elements: facial expression, cry, peripheral mobility, and/or postural tone. Additionally, despite various assessment scales, pain remains challenging to evaluate and consequently to manage in the hospital setting in general and particularly in newborns.

The AI MOTHER Neo solution is based on the technical foundation of

1. automated facial expression recognition.
2. Cry Recognition:
3. Movement and Posture Recognition:

The justification for the duration of the research is based on several factors:

1. Number of births per year: The number of births recorded at Poissy-Saint-Germain Hospital is 4,500 per year.
2. Frequency of acquisitions: Acquisitions will be made during visits on days 0, 1, 2, 3, and 4 after birth, providing a potential of 22,500 acquisitions per year.
3. Recruitment objective: The goal is to obtain 10,000 acquisitions during the research.
4. Desired recruitment/acquisition rate: A recruitment/acquisition rate of less than 15% is targeted.

Based on these data, a duration of 3 years is proposed to achieve the goal of 10,000 acquisitions. This allows for maintaining a reasonable and achievable recruitment/acquisition rate within the scope of the research.

Primary Objective Our project aims to develop a computerized clinical assessment system for newborns that takes into account the three major criteria of various clinical scales: facial expression, cry, posture, and movement.

Secondary Objectives The secondary objective is to characterize motor patterns that are specific to a given situation, particularly a pathological situation. The primary goal is to differentiate between a "normal" pattern and an "abnormal" pattern. Once a sufficient number of acquisitions have been obtained, we will attempt to identify specific patterns, meaning to make diagnoses based on computerized analysis. Patterns corresponding to early neonatal infection, anoxic-ischemic encephalopathy, brachial plexus paralysis, among others, will be identified and defined with the aim of maximizing the specificity and sensitivity of the tests.

Primary Evaluation Criterion The chosen evaluation criterion is the comparison of the computer score with the clinical examination. These scores will be described for each patient by the mean and standard deviation on normally distributed numerical parameters (median, 25th and 50th percentiles). Comparisons between the two groups will be made using paired tests.

Secondary Evaluation Criteria List the secondary evaluation criteria that address the secondary objectives.

The acquisitions consist of 2D videos at 60Hz lasting 1 to 2 minutes, taken during bathing or clinical examination of an awake, undressed child, either in a diaper or onesie. These acquisitions can be made at multiple times during the maternity stay (Days 0/1/2/3/4) for the same child and for different children. The collected data will include skeleton data where variables such as amplitude, symmetry, acceleration, limb angles, etc., will be analyzed. These various data will be combined to provide a specific movement profile and define specific patterns for a group of patients with a common characteristic.

Number of Participating Centers This is a multicenter research involving two centers located in the Paris region:

• The maternity and neonatal medicine and intensive care units at Poissy-Saint-Germain Hospital
• The Pediatric Intensive Care Unit at Raymond Poincaré Hospital, Garches. (Possible extension to a surgical intensive care unit and/or neonatology service).

Subject Identification In this research, subjects will be identified as follows: Center number (3 numeric positions) - Person's selection order number within the center (4 numeric positions) - Initial of last name - Initial of first name This reference is unique and will be retained throughout the duration of the research.

Eligibility Criteria After informing the parents or legal guardians about the objectives and procedures of the study (oral information), the protocol will be proposed to them.

Inclusion Criteria The inclusion criteria are as follows:

• Children hospitalized in the maternity ward and in the Neonatal Medicine and Intensive Care Unit at Poissy-Saint-Germain Hospital.
• Agreement of legal guardians / non-opposition. Exclusion Criteria
• Refusal of legal guardians.
• Minor parents.

Conduct of the Research The project aims to develop an automated analysis system for sound and image using computer vision in newborns in maternity units. The first phase of this project involves recording, with consent, sound and video sequences of newborns hospitalized in the maternity ward or in the Neonatal Medicine and Intensive Care Unit at Poissy-Saint-Germain Hospital. These video sequences will be analyzed by facial recognition programs developed at the R2P2 Laboratory of the Pediatric Intensive Care Unit at Raymond Poincaré Hospital in Garches and by the company OSO Ai.

Method:

The acquisition is performed on a naked or diapered, awake child lying on their back for 1 minute before bathing or clinical examination. Clinical data of patients such as sex, gestational age, birth weight, Apgar score, cord blood gases (pH and lactates), and mode of delivery will be associated with each video.

The first phase of the project involves retrieving video/audio sequences of newborns at Days 0/1/2/3/4 or beyond if hospitalization is prolonged. These sequences will focus on the child and provide a complete view including the limbs. The videos are stored on an encrypted hard drive under the responsibility of Professor Bergounioux, who will regularly extract them. These sequences will then be analyzed at the R2P2 Laboratory affiliated with the Pediatric Neurology, Rehabilitation, and Intensive Care Unit at Raymond Poincaré Hospital using modified open-source programs for facial and posture recognition as well as sound analysis. A correlation will be established with clinical evaluations conducted in parallel with the video recordings.

Population Follow-up Video/audio sequence acquisition will be conducted for newborns at Days 0/1/2/3/4 and beyond in case of continued hospitalization, either in the maternity ward or in the Neonatal Medicine and Intensive Care Unit at Poissy-Saint-Germain Hospital.

Duration of Patient Participation Patient participation is limited to the capture of images during their hospitalization.

Non-interventional research involving human subjects poses no risk to patients. Any adverse effects observed in patients participating in the research are reported by the investigators according to the local surveillance plans established within the framework of healthcare activities.

The objective is to develop a system that allows for automated analysis of the newborn on both visual and auditory levels. The primary goal is to define the "normality" of this clinical presentation, which involves acquiring a total of 10,000 video acquisitions (1 to 2 minutes each) at Days 0/1/2/3 and Day 4 for newborns hospitalized in the maternity ward or in the Neonatal Medicine and Intensive Care Unit.

These films will be obtained with parental consent and solely intended for research and development of the automated system AI MOTHER Neo under the responsibility of Professor Bergounioux.

The analysis aims to establish classification algorithms using neural networks that differentiate between a "normal examination" and an "abnormal examination."