INTRODUCTION: Chronic Widespread Pain (CWP), initially a fibromyalgia criterion, is now independently recognized in ICD-11, advancing chronic pain assessment [1,2]. CWP(MG30.01) is defined as pain lasting over three months in at least three body quadrants across four of five regions, including the axial skeleton[3]. Pain must be associated with emotional distress or interference with daily activities and cannot be explained by another chronic pain condition. This study aims to validate the ICD-11 anatomical criteria for CWP by developing a machine learning model for automated body map analysis to identify pain distribution subtypes in chronic low back pain (cLBP) patients.This approach establishes a basis for integrating body map phenotyping data into a multi-modal framework, allowing analysis of pain distribution patterns alongside other phenotypic data from comprehensive assessments in a deep phenotyping study on cLBP patients at UCSF (COMEBACK trial) [4].
METHODS: This study uses data from the UCSF-COMEBACK trial, part of the BACPAC consortium, with patients experiencing LBP for over six months without a clear cause. An automated workflow processed body map images to quantify pain regions. The original body map outline was subtracted from the patient drawing to extract pain locations. Images were cropped, resized, and aligned by predefined quadrants and regions for both anterior and posterior views, preserving aspect ratios (Fig.1 and 2). A mask-based segmentation technique identified and quantified pain by counting pixels in reported pain areas. Connected component analysis isolated distinct pain clusters within the masks (Fig.3). Pixel counts were compiled for statistical analysis. Patients were classified as having CWP if they reported pain in more than four out of five regions, including at least one axial. Additionally, pain had to be present in over three of four quadrants. Regions were considered painful if reaching a cutoff of five pixels. Training was conducted on 297 patients, with validation on a random sample of 42, comparing automated CWP classifications against manual assessments to evaluate accuracy and consistency (Fig. 4). Remaining patients were used to refine segmentation techniques.
RESULTS: A total of 339 patients were included, with validation showing approximately 76% accuracy in identifying pain regions. Among the 10 misidentified cases, 90% had fewer pixels in extracted pain regions compared to the original template, and 40% were affected by boundary issues due to template edges.
DISCUSSION: The inclusion of CWP in ICD-11 represents a significant advancement in recognizing chronic pain. Although some limitations were present, the tool demonstrated promising accuracy in identifying pain regions. This study highlights the potential of automated body map phenotyping for comprehensive CWP analysis, with plans to integrate body map data with additional COMEBACK data to enable broader exploration of pain patterns, allowing for more consistent, scalable, and objective identification of pain regions compared to manual readings. Examining concordance and discordance among phenotypic domains and their correlations with clinical data, including imaging, quantitative sensory testing, and biomarkers, aims to enhance objective pain assessment and deepen understanding of pain mechanisms.
Fig.1 Body map quadrants.
Fig.2 Body map regions.
Fig.3 Workflow for body map processing.
Fig.4 Validation workflow.