Page 84 - AIH-1-4

Artificial Intelligence in Health Robotics and Vivaldi AI for ALS assessment

Bland–Altman plot is a graphical method to compare two we integrated this questionnaire into the Sanbot Elf robot,
measurements techniques, 31,32 also providing a quantitative leveraging the advanced capabilities of the Vivaldi AI system
estimate of how closely the values from two measurements to create an adaptive and user-friendly interface for patients.
lie. In this context, we reported both the bias (mean of the The ALSFRS-R questionnaire, embedded in the chest
33
differences between the two methods) and the 95% limits tablet interface of the Sanbot Elf, comprises 12 items that
of agreement (LOA). The ICC provides a single measure of assess four key domains: bulbar function, fine motor
the extent of agreement between the two methods, and function, gross motor function, and respiratory function.
33
we considered the guidelines recommended by Koo and Each item is scored 0 (complete loss of function) to 4
Li for interpreting the results. In detail, basing on the (normal function), enabling a thorough evaluation of the
34
95% confident interval of the ICC estimate, values <0.5, patient’s abilities.
between 0.5 and 0.75, between 0.75 and 0.9, and >0.90 are
indicative of poor, moderate, good, and excellent reliability, In a hospital setting, the Sanbot Elf administered the
respectively. Moreover, to longitudinally quantify the ALSFRS-R questionnaire through an interactive process
34
heterogeneity in rates of decline in ALSFRS-R between tailored to the specific needs of ALS patients, with a typical
individuals, we calculated the coefficient of variation (CoV). duration of approximately 10 – 15 min. The Vivaldi AI
The CoV was defined as the between-patient standard system guided this administration, ensuring a smooth and
deviation of slope divided by the mean rate of change, and adaptive experience. Patients responded to each question
a lower value indicates both a less variation among patients, using a simple “yes” or “no” format, which minimizes
which could positively affect sample size calculations, and an ambiguity and ensures clarity in their responses, and the
increase of the sensitivity in detecting disease progression. Vivaldi AI system dynamically adjusted the sequence of
Finally, a mixed-effects linear regression with ALSFRS-R questions based on the patient’s responses.
(total score and subgroups, separately) as the outcome, and Sanbot Elf offers multiple interaction modes to
fixed-effects for time, approach (human operator vs. robotic accommodate the diverse needs of ALS patients. Indeed,
operator), and an interaction between time and approach. In patients with limited mobility but who can speak
detail, the fixed effect for the interaction between approach were able to interact verbally with the robot. For those
and time was used as an assessment of bias in the human- experiencing dysarthria or hypophonia, the touchscreen
based score compared to the robot-based score. Considering interface provided an alternative method for responding to
the psychological assessment, a paired t-test was used to questions. This dual-mode interaction ensured inclusivity
evaluate the impact of the questionnaire’s administration and accessibility for all patients.
by the robotic operator on patients’ anxiety, comparing the
state anxiety level pre- and post-administration. Finally, the In addition, patients could use vocal commands to
correlation between the bias in evaluating the ALSFRS-R request repetitions of questions or take brief pauses, a
between the two methods and the “openness to experience” feature facilitated by the Vivaldi AI system (this could
dimension of the BFI was evaluated using Spearman’s rank cause an increase in the time required to administer the
correlation coefficient. Regarding the qualitative analysis questionnaire, without negatively influencing the operator’s
collected from the observational grid and the semi- burden). This enhanced comfort and ease of use, catering
structured interview, frequency and percentage were used to to patients who might need more time or clarification.
investigate the emotional states and the patients’ perception On completion of the questionnaire, the Vivaldi AI
about the experience. Specifically, considering the analysis system automatically calculated the final ALSFRS-R score,
of the semi-structured interview, a cluster’s identification promptly providing results. This score is useful for ongoing
was obtained by analyzing the answers given by patients monitoring and assessment of the patient’s functional
for the categories of each thematic area. All statistical tests status and disease progression. The automated scoring
were two-tailed, and P < 0.05 was considered statistically process ensures accuracy and consistency, minimizing the
significant. All the statistical analyses were performed using risk of human error.
SAS 9.3 (SAS Institute, Inc, Cary, NC) software. For the text
analysis, the Atlas.ti software was used. 5.2. Agreement analysis between human and
robotic ALSFRS-R assessment
5. Results
A group of 28 ALS patients participated in this study, with
5.1. Design and implementation of the ALSFRS-R in a median age at evaluation of 62.37 years (range: 53.88
Sanbot Elf – 68.43) and a male-to-female ratio of 2.11. Descriptive
The ALSFRS-R questionnaire is an essential tool for characteristics of the ALS cohort, including demographic
evaluating the functional status of ALS patients. In our study, and clinical features, are summarized in Table 1.

Volume 1 Issue 4 (2024) 78 doi: 10.36922/aih.3732