Refining mUltiple Artificial intelliGence strateGies for Automatic Pain Assessment Investigations: RUGGI Study

Not Applicable

Recruiting

• Conditions: Chronic Pain; Cancer Pain; Neuropathic Pain; Pain Assessment

• Registration Number: NCT07038434
• Lead Sponsor: Valentina Cerrone

Brief Summary

This single-center, non-profit, observational-interventional study aims to develop artificial intelligence (AI) models for the automatic assessment of chronic pain (APA - Automatic Pain Assessment). The study will enroll adult patients with chronic pain of various origins (oncologic and non-oncologic). Participants will undergo multidimensional evaluations that include clinical assessments, self-report questionnaires, bio-signal collection (e.g., EEG, EDA, HRV, GSR, PPG), and facial expression analysis via infrared thermography and video recordings.

The primary objective is to calibrate and test machine learning and deep learning models to recognize and predict the presence and severity of pain using multimodal data inputs. Secondary objectives include evaluating the effectiveness of pain treatments, assessing quality of life, and developing a standardized APA dataset for future research.

All data collection procedures are non-invasive and safe, and include tools like wearable sensors and standardized neurocognitive tests. The study is approved by the Italian Ethics Committee (Comitato Etico Territoriale Campania 2) and complies with GDPR and EU AI regulations.

Detailed Description

This study, titled "Refining mUltiple artificial intelliGence strateGies for automatic pain assessment Investigations" (RUGGI), explores the integration of AI in chronic pain evaluation. Pain is a multidimensional and subjective experience, and conventional assessment methods often rely solely on self-reported scales. This introduces the risk of over- or under-treatment. To overcome this limitation, the study leverages multimodal data-including physiological signals, facial expressions, and linguistic analysis-to build models capable of objectively assessing pain intensity and characteristics.

The primary aim is to calibrate predictive models (e.g., Support Vector Machines, Random Forest, Convolutional Neural Networks, YOLO architectures, and MLPs) that can recognize pain patterns using supervised and unsupervised learning. Bio-signals (EEG, HRV, GSR, EMG), infrared thermography (HIRA system), and prosodic-linguistic features will be analyzed. Data will be collected during structured timepoints: baseline (rest), Stroop test execution, and follow-up.

Patients are recruited based on chronic pain diagnosis per IASP and ICD-11 criteria. Inclusion criteria include age ≥18 and informed consent. The study foresees a target enrollment of approximately 200 patients within 6 months. Data will be processed following a rigorous AI pipeline, including preprocessing, feature extraction, dimensionality reduction, and cross-validation (k-fold with grid search optimization). Outcome measures include the Area Under the Curve (AUC), sensitivity, specificity, F1 score, and model explainability (via SHAP, LIME).

Secondary outcomes include assessing patient-reported quality of life, evaluating analgesic strategies, and generating a public-use APA dataset. All procedures are compliant with Good Clinical Practice (GCP), GDPR, and EU Artificial Intelligence Act (Reg. 2024/1689). The study is conducted at the University Hospital "San Giovanni di Dio e Ruggi d'Aragona" in Salerno, Italy.

Study Timeline

• Study Start: 2025-05-06
• Study First Submitted: 2025-05-19
• Status Verified: 2025-06
• Study First Submitted QC: 2025-06-17
• Last Update Submitted: 2025-06-17
• Study First Posted: 2025-06-26
• Last Update Posted: 2025-06-26
• Primary Completion: 2025-12
• Study Completion: 2026-01-01

Recruitment & Eligibility

• Status: RECRUITING
• Sex: All
• Target Recruitment: 200

Inclusion Criteria

• Adults (≥18 years old) with chronic pain, defined according to IASP and ICD-11 as pain that persists or recurs for more than three months.
• Diagnosed with either:
• Chronic primary pain (e.g., fibromyalgia, irritable bowel syndrome, chronic headaches)
• Chronic secondary non-cancer pain (e.g., low back pain, osteoarthritis, post-surgical pain)
• Chronic cancer-related pain (due to cancer or its treatment)
• Ability to understand the study procedures and provide written informed consent.

Exclusion Criteria

• Current treatment with psychotropic drugs or presence of active psychiatric disorders (e.g., psychosis, major depression).
• Known history of alcohol or substance abuse.
• Pregnancy or breastfeeding.
• Age under 18 years.
• Inability to provide informed consent (e.g., due to cognitive impairment).

Study & Design

• Study Type: INTERVENTIONAL
• Study Design: SINGLE_GROUP

Primary Outcome Measures

Name	Time	Method
Accuracy of AI models in classifying chronic pain	From Day 0 (baseline) to Day 30 (follow-up)	Accuracy will be calculated to evaluate how well supervised machine learning and deep learning models can correctly classify the presence of chronic pain using multimodal data (e.g., biosignals, facial thermography, video, and audio).
Sensitivity of AI models in classifying chronic pain	From Day 0 to Day 30	Sensitivity (true positive rate) will be computed to determine the model's ability to correctly identify patients experiencing chronic pain.; Unit of measure: Sensitivity (%)
Specificity of AI models in classifying chronic pain	From Day 0 to Day 30	Specificity (true negative rate) will be computed to assess the model's ability to correctly identify patients who are not experiencing chronic pain.; Unit of measure: Specificity (%)
Precision of AI models in classifying chronic pain	From Day 0 to Day 30	Precision (positive predictive value) will be calculated to assess the proportion of correct positive predictions among all positive classifications.; Unit of measure: Precision (%)
F1-score of AI models in classifying chronic pain	From Day 0 to Day 30	F1-score, the harmonic mean of precision and sensitivity, will be used to assess overall model performance, especially in the presence of class imbalance.; Unit of measure: F1-score (numeric value)
AUC-ROC of AI models in classifying chronic pain	From Day 0 to Day 30	The area under the receiver operating characteristic curve (AUC-ROC) will be used to evaluate the model's ability to discriminate between pain and no-pain conditions across thresholds.; Unit of measure: AUC-ROC (numeric value from 0 to 1)

Secondary Outcome Measures

Name	Time	Method
Change in Patient Global Impression of Change (PGIC) score	From Day 0 to Day 30	This outcome will measure patients' perceived improvement in their condition using the PGIC scale.; Unit of measure: Score on a 7-point Likert scale (1 = No change to 7 = Very much improved)
Change in Brief Pain Inventory (BPI) interference score	From Day 0 to Day 30	This outcome will measure how much pain interferes with daily functioning, using the BPI interference subscale.; Unit of measure: Score from 0 (no interference) to 10 (complete interference)
Correlation between analgesic treatments and pain intensity (NRS)	From Day 0 to Day 30	The outcome will assess the correlation between the type and frequency of analgesic treatments and changes in pain intensity, measured with the Numeric Rating Scale (NRS).; Unit of measure: Pearson correlation coefficient (r), NRS scores from 0 to 10

Trial Locations

Locations (1)

Azienda Ospedaliera Universitaria San Giovanni di Dio e Ruggi d'Aragona; 🇮🇹 Salerno, Italy