5150. MRI-based Musculoskeletal Infection-Reporting and Data Systems (MSKI-RADS): Multicenter and Multireader Validation
Authors * Denotes Presenting Author
  1. Angela He *; UT Southwestern Medical Center
  2. Karim Salhadar; UT Southwestern Medical Center
  3. Mina Guirguis; UT Southwestern Medical Center
  4. Flavio Duarte Silva; UT Southwestern Medical Center
  5. William Morrison; Thomas Jefferson University Hospital
  6. Yin Xi; UT Southwestern Medical Center
  7. Avneesh Chhabra; UT Southwestern Medical Center
A standardized guideline and reporting system can improve the evaluation and characterization of the MRI diagnosis of musculoskeletal (MSK) infections for improved multidisciplinary communications and timely management. This study aims to develop and validate a MSK infection classification and MRI-based scoring system titled "musculoskeletal infection-reporting and data systems (MSKI-RADS) and validate it with multireader assessment.

Materials and Methods:
This was an IRB-approved, multicenter validation study with waiver of patient consent. Expert MSK radiologists designed a MSKI-RADS system using the terms described in the SSR white paper on MSK infection and outlined management recommendations accordingly. Scoring categories were created to account for the spectrum of infection severity based on different tissue compartments: 0 - incomplete imaging, I - negative for infection, II - superficial soft tissue infection, III - deeper soft tissue infection, IV - possible osteomyelitis (OM), V - highly suggestive of OM, VI - known OM, NOS - nonspecific bony lesions unrelated to infection. Subcategories in class V and VI further characterized the severity and/or the extent of the infection as V and VI A, B, or C. Validation was performed by 20 MSK readers from 12 institutions with experience levels ranging from 1 - 28 years post-musculoskeletal fellowship. Readers received initial online training on the scoring system. A set of randomly selected cases with complete imaging and spectrum of proven musculoskeletal infections was then presented to the readers, blinded to final diagnosis. Interreader agreement (ICC and Conger’s kappa) and accuracy testing was performed.

Among a total of 208 cases, there were 20 negative, 34 superficial soft tissue infections, 35 deeper soft tissue infections, 30 possible OM, 35 highly suggestive of OM, 18 known OM, and 36 NOS cases. Substantial interreader agreement was observed using MSKI-RADS among the 20 readers (ICC: 0.7 [0.65, 0.74]). There was no correlation between reader experience level and reader accuracy (<em>p</em> > 0.05). The average reader accuracy for noninfectious cases (class I and NOS), soft tissue infection (II and III), and bony involvement with infection (IV and V) were 82.3%, 75.6%, and 84.4% respectively. The average reader accuracy for superficial soft tissue infection (II), deeper soft tissue infection (III), and possible osteomyelitis (IV) were 85.6%, 88.1%, and 88.1% respectively. For osteomyelitis (V) and known osteomyelitis cases (VI), the average reader accuracy was 87.2% and 97.5%, respectively.

The MSKI-RADS classification system is accurate and reliable across the spectrum of MSK infections and among MSK-trained readers from multiple centers with differing experience levels. This system can be tested in future studies with general radiologists for broader applicability. This will be a living document and can be updated as more data becomes available in future.