This was a retrospective, HIPAA compliant, cross-sectional multi-institutional and multi-reader study involving three institutions. Anonymized MRI studies (with all 18 HIPAA identifiers stripped) were presented to the readers with proven cases randomly extracted from different institutions as part of retrospective Institutional review boards e.g. the experimental protocol was approved by UT Southwestern institutional IRB (STU 112017-003). As per this IRB, the informed consent was waived by UTSW ethics committee. No data use agreement is required for such studies without HIPAA information.
Inclusion criteria: Musculoskeletal soft tissue tumors of upper and lower extremities with complete MR imaging sets (as defined below) of adipocytic (fat containing on MRI or histology), T2-hyperintense tumors and tumor-like lesions (hyperintense than muscle signal) and T2-hypointense masses and tumor-like lesions with histology proof via biopsy and/or final post-surgical histology, and the exclusion criteria were incomplete imaging set and lack of final clinical diagnosis, e.g. lack of arthroscopy for parameniscal cyst or final clinical diagnosis of Gout.
Development of Soft-tissue Tumor Reporting And Data System (ST-RADS) consensus document
ST-RADS consensus document was created by reviewing WHO document, the consensus opinion of tumor imaging experts from all three institutes, input from clinicians and expert methodologist from the primary institution (Table. 1). WHO classification of soft tissue tumors was used as a guide to place a spectrum of commonly encountered histologies in various categories of ST-RADS (Table. 2). As indicated in Table. 1, all soft tissue tumors were classified into one of the ST-RADS categories 0-VI as outlined in Table.3 and were evaluated for study purposes.
ST-RADS 0 was used in the event of a non-diagnostic study / incomplete imaging, and further imaging is required.
A complete MR imaging study is defined as the one with full tumor coverage in each of these imaging sequences in at least one plane: T1W imaging (T1WI), fluid sensitive sequence (fat suppressed T2WI or T2WI, or inversion recovery), and post-contrast fsT1WI. An incomplete MR imaging study is defined as the absence of one of the above-described sequences [55, 58, 59].
A recent examination with an additional complement of MRI sequences may suffice as complete MR imaging if obtained within two weeks of an incomplete imaging set. Diffusion-weighted imaging (DWI) is an emerging modality and is not universally used [9, 23, 46]. Apparent diffusion coefficient (ADC), if available during interpretation, can be used as a supplemental finding for the diagnosis [3, 9, 23, 42, 46, 49].
ST-RADS I was used if no lesion is identified, and no further follow-up is needed.
ST-RADS II was used when the lesion is definitely benign, and no follow-up needed. Representative masses include simple lipoma with a uniform fat signal on all sequences with complete suppression on fat saturation or inversion recovery images and no appreciable enhancement, can be upto 10cm [8, 17, 19], completely calcified hypointense lesions as also confirmed with radiographs or CT imaging, or fluid intensity (markedly hypointense on T1WI and markedly hyperintense on T2W) mass connected to a joint or bursa with no intravenous contrast enhancement [32, 37]. Other masses include those with classic imaging features and spatial location of a benign tumor, such as plantar fibroma, fat necrosis [48], elastofibroma dorsi of chest wall [44, 49], fibrolipoma of nerve [30, 31, 53], hemangioma or venous malformation with phlebolith (s), especially with additional radiographic demonstration [12, 60], peri-articular gouty tophus with cortical erosions (especially if radiographs are available) in the setting of elevated serum uric acid levels, Morel-Lavallée lesion [52], parameniscal or paralabral cyst or geyser phenomenon, venous or lymphatic malformation [16], sarcoid or rheumatoid nodule in classic locations with a known history of the respective systemic disease, Morton’s neuroma, and thrombosed vein or artery. ST-RADS II lesions usually show no enhancement or thin (< 2 mm) peripheral and/or septal enhancement on the post-contrast study with no significant diffusion restriction. Variable enhancement may be seen with otherwise classic lesions, such as plantar fibroma, Morton’s neuroma, elastofibroma dorsi, hemangioma and arthritis related nodules. DWI, if obtained, exhibits moderate-marked hyperintensity of the lesion on both DWI and ADC images, with a mean ADC value>1.5-3.0 x10-3 mm2/s [9].
ST-RADS III was used for probably benign masses with less than or equal to a 2% chance of malignancy. These masses are recommended to be followed until two years or until the lesion spontaneously resolves or significantly regresses. Representative masses include fluid intensity lesion adjacent to a joint or bursa, possibly a septated ganglion cyst, or fibrous tissue signal lesions in relation to fascia and muscles, e.g., classic palmar or plantar fibromatosis [29] (moderately-markedly hypointense on T1W and T2W imaging), or intramuscular lesion such as a myxoma [6, 40]. Intra-muscular myxoma can be a pathognomonic diagnosis [61] but may show heterogeneous enhancement, that is why it is placed in ST-RADS III and not II or IV. Plantar fibroma is placed in ST-RADS II as it is a classic benign diagnosis with a pea-shaped enhancing nodule in relation to central cord of plantar fascia. Fibromatosis is however placed in ST-RADS III as these can be larger, sheet-like, multifocal, locally aggressive (rarely metastasizing), and can mimic low grade sarcoma [34]. In addition, ST-RADS guide is not intended to match a particular histology with a particular ST-RAD grade as the histology wouldn’t be known prospectively but is envisioned as a system that can be of practical use and which can be tested prospectively refined as more experience is garnered over the years. Other tumors or tumor-like lesions include those with typical imaging features, e.g., lipoma with metaplastic calcification or ossification, myositis ossificans (history of trauma, internal hemorrhage, marbled muscle-like appearance of the mass, and developing peripheral calcification on the corresponding radiograph or CT), benign peripheral nerve sheath tumors [30, 53] (target or fascicular sign, absent peritumoral edema or necrosis, length < 5 cm, underlying schwannomatosis, absent rapid growth or new or sudden neurological deficit or excessive pain), synovial chondromatosis, tenosynovial giant cell tumor (TSGCT), Desmoid, and angiolipoma [19]. Post-contrast imaging of such lesions may exhibit no enhancement or thin (< 2 mm) peripheral and/or septal enhancement, such as a ganglion cyst or bursitis, or variable enhancement with other lesions. DWI, if obtained, exhibits moderate-marked hyperintensity of the lesion on DWI and mild-moderate hyperintensity on ADC images, with generally a mean ADC value = 1.2-2.0 x 10-3 mm2/s. Caution needs to be exercised with certain lesions, such as TSGCT and myxoid lesions [34]. TSGCT typically shows hypointense to moderately hyperintense DWI signal, and mean ADC can be low, varying from 0.8-1.3 x 10-3 mm2/s [3]. Both benign and malignant myxoid tumors commonly exhibit high ADC values [36].
ST-RADS IV was used for potentially malignant, but indeterminate tumors, and the suspicion for malignancy is more than 2% and less than 50%. The recommendation is tissue biopsy or short-term follow up in 4-6 weeks, and regular interval follow-up for upto 2 years. Tumors in this category may exhibit mixed intensity on MR imaging, a solid appearance, however less than 5 cm maximum length, or a lipomatous lesion with multiple septations without a solid focal nodule or myxoid change. Representative tumors include atypical lipomatous lesion [19, 26], solitary fibrous tumor, and Gardner fibroma [29]. On post-contrast imaging, there is variable enhancement. DWI, if obtained, exhibits moderate-marked hyperintensity of the lesion on DWI and mild-moderate hyperintensity on ADC images, with generally a mean ADC value > 1.1 x 10-3 mm2/s [3, 9, 62]
ST-RADS V was used for probably malignant tumors where tissue diagnosis is recommended. The probability of malignancy in this group of tumors is 50% or more. On MRI, these tumors exhibit mixed intensity and a solid mass, or a lipomatous lesion containing multiple thick septations or solid nodule (s), and/or myxoid changes or a lipid poor T2 hyperintense enhancing mass. Representative tumors in this category include malignant peripheral nerve sheath tumor [30, 53], synovial sarcoma [24, 50], undifferentiated pleomorphic sarcoma, myxofibrosarcoma [25], melanoma, and lymphoma [43]. These lesions show variable and solid enhancement with diffusion restriction and mean ADC values =< 1.1 x 10-3 mm2/s[9]. As with other myxoid tumors, myxoid sarcomas typically have areas of higher mean ADC [42].
ST-RADS VI was used for a known biopsy-proven malignancy or recurrent malignancy in the tumor bed prior to definitive therapy. Recommendations include surgical excision or further treatment as clinically appropriate. Solid nodule or residual/growing mass with imaging features like the pre-intervention lesion in the tumor bed suggests tumor recurrence. Post-contrast imaging typically shows solid nodular enhancement or enhancement similar to the tumor prior to intervention in such cases [10, 15, 39]. DWI, if obtained, shows diffusion restriction and mean ADC <1.1x10-3 mm2/s or similar to that of the pre-intervention tumor [15].
The consensus document was edited, shared electronically among experienced radiologists with sarcoma imaging, and discussed during conference calls. Once agreed upon, multi-reader testing was performed at all three sites. Instead of reviewers guessing qualitatively during read-outs and mentioning likelihood of malignancy on a Likert scale as unlikely, possible, probable, highly likely, etc., quantitative numbers were presented to the reviewers to remind them about the categories as has been validated in the BI-RADS system and how they would have practically reported such studies, e.g. is it significantly indeterminate with still less than 50% probability that one could present a choice of tissue sampling or a short-term follow-up versus more than 50% chance of malignancy that one should obtain tissue sampling. The goal of ST-RADS is a dynamic guideline document that may be refined as more evidence is collected over time.
For validation and testing purposes, commonly encountered tumors in all three described categories were used: adipocytic tumors, T2-hyperintense and T2-hypontense tumors encompassing a mixture of a wide spectrum of common and uncommon histologies as identified from the WHO classification [54]. The hyperintense or hypointense categories were based on predominant tumor appearance on T2W imaging, meaning at least 50% or more of tumor appeared hyperintense or hypointense, respectively.
Data collection and rating procedures
A random sample of soft tissue tumors with ‘complete’ MR imaging sets of adipocytic and T2-hyperintense tumors and tumor-like lesions with proven histopathological diagnoses were collected and shared among the three institutes using Microsoft PowerPoint presentations (Office 365, Microsoft, Seattle, WA) via online meetings. The data sets were anonymized, and the readers of each site were presented with tumors from the other site. The tumors were displaced in the entirety with most-representative images. The readers were blinded to the final histological diagnosis and the interpretation of the other readers. A musculoskeletal fellow at UTSW controlled all data and compiled all interpretations in an Excel database (Windows 10, Microsoft, Redwood, WA). DWI and ADC images were available as supplemental images in a few cases, specifically four T2-hyperintense and one adipocytic tumors.
The principal investigator conducted a training session on three separate occasions to standardize the understanding of the ST-RADS consensus document before independent scoring. A total of 200 soft tissue tumors (100 adipocytic tumors, 50 T2-hyperintense tumors, and 50 T2-hyponitense tumors) were evaluated (Table. 3). Eight Musculoskeletal radiologists evaluated the studies with their attending level experience ranging from 2 years post-fellowship to more than 30 years of interpreting MR imaging of soft tissue tumors (Table. 4).
Statistical Analysis
Intraclass correlation coefficient (ICC (2,1)) with a mixed-effects model was used to assess inter-reader agreement and the reliability of the guideline system. Median ST-RADS from readers use to assess diagnostic performance. Areas under the receiver operating curve (AUC) were calculated. Sensitivity and specificity were also calculated with I/II/III as benign and IV/V as malignant with final histology diagnosis bases on biopsy and/or surgery serving as the reference standard. All analyses were done in R 4.0.2 (Vienna, Austria) by an expert statistician from the primary institution.
The following interpretation of ICC was used: Excellent Agreement: 0.75 - 1.00; Good Agreement: 0.60 - 0.75; Fair Agreement: 0.40 - 0.60; and Poor Agreement: < 0.40 [11].