Soft tissue sarcomas (STS) are rare malignancies of mesodermal origin, accounting for just 1% of all malignant tumours. Despite their rarity, they carry significant mortality: the American Cancer Society estimated 13,460 new STS cases and 5,350 deaths in 2021 alone. STS most commonly arise in the extremities or trunk, and they typically present as a soft tissue lump, often first encountered by a primary care physician or general practitioner. This deceptively simple presentation means delays in diagnosis are common, making imaging protocols and specialist referral pathways critically important.
Management and prognosis of STS depend on a constellation of factors: tumour size, anatomical location, histologic subtype, TNM stage, histological grade, patient age, and comorbidities. The backbone of treatment is limb-sparing surgical resection, often combined with radiotherapy administered either before surgery (neoadjuvant) or after (adjuvant). Chemotherapy has a more contested role and remains under active debate in the literature. Given the heterogeneity of STS subtypes, which number more than 50 distinct histological entities, a one-size-fits-all approach to imaging is not feasible. This review by Vibhakar and colleagues (Leicester Royal Infirmary, Kettering General Hospital, and Royal Orthopaedic Hospital Birmingham) synthesises current best practices across all major imaging modalities and highlights recent advances including diffusion-weighted MRI, dynamic contrast-enhanced MRI, whole body MRI, hybrid PET/MRI, and early applications of artificial intelligence.
The article explicitly restricts its scope to extremity STS and excludes retroperitoneal sarcomas, which have different anatomical constraints, surgical considerations, and imaging requirements. This focused approach allows a thorough treatment of the imaging considerations most relevant to the large majority of STS cases. The review is structured around each imaging modality in turn, making it a practical reference for clinicians deciding which technique to deploy at each stage of a patient's diagnostic and treatment journey.
Ultrasound is almost universally the first imaging study ordered for a soft tissue lump. It is readily accessible, does not use ionising radiation, and allows real-time assessment of lesion depth, vascularity, and character. According to the review, ultrasound can correctly diagnose a benign lesion with an accuracy of 80-90%, and aggressive lesions can be identified with a sensitivity and specificity both around 84%, based on multiple published studies. Specific sonographic features guide triage: lesions greater than 46 mm in longest diameter, hypervascular with multiple peripheral poles, heterogeneous with ill-defined margins, and located deep to the deep fascia are considered aggressive or indeterminate and warrant specialist referral. Conversely, small subcutaneous avascular lesions with well-defined margins are likely benign.
Diagnostic utility by application: Ultrasound achieves 96% specificity for diagnosing superficial lipomas, making it highly reliable in ruling in this common benign entity without further imaging. For recurrence surveillance, the modality demonstrates an overall sensitivity and specificity of 83-84% and 93-94% respectively for detecting local tumour recurrence. However, no published study has demonstrated superiority of ultrasound over MRI in recurrence surveillance when the two are compared head-to-head. Doppler interrogation aids in distinguishing vascular recurrent tumour from avascular fibrous scar tissue, though hypovascular recurrence can mimic benign appearances and create false reassurance.
Ultrasound-guided biopsy: Beyond diagnosis, ultrasound plays an important procedural role. Real-time visualisation during biopsy allows direct targeting of viable, vascular tumour tissue, with avoidance of adjacent neurovascular structures, necrotic regions, and cystic areas. This is particularly valuable given that the diagnostic yield of biopsy is critically dependent on sampling the right tissue. One notable strength is ultrasound's ability to differentiate myxoid lesions, which can appear falsely cystic on MRI, from truly cystic structures, preventing misclassification.
Limitations: Key limitations include operator dependence (making serial comparisons unreliable), inability to adequately characterise lesions for planning surgical re-excision of recurrent sarcoma, and restricted field of view. Despite these shortcomings, ultrasound remains the cornerstone of initial STS assessment and has a clear cost-effective role in triaging lumps and preventing unnecessary downstream investigations.
Computed tomography plays a more limited but clearly defined role in STS management. For locoregional staging, MRI is superior due to its far better soft tissue contrast resolution and multiplanar capability. However, CT fills specific niches that MRI cannot occupy. CT features associated with malignant behaviour include margin irregularity, infiltration into adjacent organs, calcification, necrosis, and hypervascularity. For assessment of osseous involvement, including cortical destruction, endosteal reaction, and periosteal changes, CT is considered equivalent or even superior to MRI in some scenarios, particularly where bone involvement is the primary question.
Staging and subtype-specific indications: STS predominantly metastasises haematogenously to the lungs, making chest CT central to staging and surveillance. Specific subtypes carry additional metastatic sites requiring dedicated imaging: abdominal CT for myxoid liposarcoma (abdominal metastases in up to 12.1% of cases) and for myxofibrosarcoma (adrenal glands, mesentery); brain CT for alveolar soft part sarcoma, clear cell sarcoma, and angiosarcoma. Regional lymph node metastases are uncommon overall but specifically associated with epithelioid and clear cell sarcomas, warranting CT staging of nodes in those subtypes.
Surveillance schedules: The American College of Radiology Appropriateness Criteria recommend non-contrast chest CT every 3-4 months for the first 2-3 years after treatment in high-risk STS patients, followed by every 6 months up to 5 years, then annually thereafter. For low-grade STS, chest CT every 6-12 months is recommended. These structured surveillance intervals reflect the temporal distribution of STS recurrence and metastasis risk.
CT-guided biopsy: When lesions are too deep for safe ultrasound access, or when adjacent critical structures cannot be reliably visualised sonographically, CT guidance is preferred. A notable example is biopsy of lesions in the posterior intercondylar notch of the knee, where precise avoidance of the popliteal neurovascular bundle requires the spatial clarity of CT. PET/CT adds further value when initial CT biopsy yields non-diagnostic material, as it can identify metabolically active peripheral tumour regions most likely to yield diagnostic tissue.
MRI is the definitive imaging modality for locoregional STS evaluation, offering superior soft tissue contrast resolution without ionising radiation. Standard STS MRI protocols include T1-weighted (T1W), T1 with fat suppression (T1FS), T2-weighted (T2W), diffusion-weighted imaging (DWI), and optionally dynamic contrast-enhanced (DCE) sequences. Each sequence interrogates different tissue properties: T1W characterises fat content and anatomical relationships; T2W reveals fluid and oedema; gadolinium-enhanced sequences assess vascularity and tissue planes. Contrast-enhanced MRI can characterise cystic versus solid lesions, assess vascular invasion, and illuminate tissue plane relationships critical for surgical planning.
Diagnostic features and their quantitative thresholds: The highest sensitivity for detecting malignancy on MRI comes from absence of low signal intensity on T2W imaging combined with a mean diameter greater than 33 mm and inhomogeneous signal on T1W imaging. The highest specificity is achieved when necrosis, bone or neurovascular involvement, or distant metastasis is identified, or when mean diameter exceeds 66 mm. STS commonly grow within anatomical compartments, form pseudo-capsules, and do not infiltrate adjacent structures until advanced stages, creating falsely reassuring morphological appearances. As a result, there is significant overlap between benign and malignant appearances, and the authors emphasise that one should have a low threshold for proceeding to biopsy rather than relying solely on MRI characterisation.
Post-operative surveillance: MRI detects more than one third of local STS recurrences before they become clinically apparent, establishing it as the cornerstone of post-treatment surveillance. Distinguishing recurrent tumour from post-surgical seroma, haematoma, inflammation, and fibrosis is a recognised challenge on conventional T1W and T2W sequences. The complete absence of fluid signal in the surgical bed is a highly specific indicator of no recurrence, but this is infrequently observed. Recurrence may manifest as architectural distortion on T1W, or as mass-like or nodular enhancement after intravenous contrast. Important exceptions include undifferentiated pleomorphic sarcoma and myxofibrosarcoma, which may recur as subtle plaque-like tails of tumour that are easily overlooked.
Whole body MRI (WBMRI): Using STIR and T1W sequences, WBMRI is valuable for detecting extent of soft tissue and bone metastases. Its strongest evidence base is in myxoid liposarcoma, where it identifies extra-pulmonary metastases that CT misses: one study reported sensitivity of 80-84.6% and specificity of 97-98.9% for soft tissue and bone metastases. WBMRI directly alters management and is now recommended at diagnosis of relapse to confirm whether only solitary metastasis is present before metastasectomy. In the paediatric setting, WBMRI is particularly favoured because it avoids ionising radiation and simultaneously evaluates bone marrow, soft tissues, and solid viscera, reducing the total number of investigations required.
Dynamic contrast-enhanced MRI (DCE-MRI) and diffusion-weighted imaging (DWI) are functional MRI techniques that go beyond the anatomical information provided by conventional sequences. DCE-MRI evaluates early enhancement kinetics of gadolinium contrast, capturing the microcirculatory behaviour of tumour tissue in ways that static post-contrast imaging cannot. Viable tumour contains tissues with increased vascularity, high capillary permeability, and rapid perfusion, producing rapid early gadolinium uptake approximating arterial enhancement curves. Non-viable tumour and post-treatment inflammation exhibit slower, delayed enhancement because their capillary resistance is higher and permeability lower. This differential enhancement timing allows DCE-MRI to categorise tissues within a complex tumour bed.
Quantitative impact on recurrence detection: The performance data for DCE-MRI in post-surgical STS surveillance are striking. Adding a DCE sequence to the standard protocol improves MRI recurrence detection from 27% to 97%, a transformation in diagnostic yield. Furthermore, DCE-MRI reduces the false-positive rate from 48% to 3.2%. Given that STS post-surgical surveillance routinely includes contrast-enhanced MRI, adding a DCE sequence requires no additional contrast administration, only additional acquisition time. The clinical implications of this improvement in specificity are substantial: unnecessary biopsies and patient anxiety from false-positive surveillance findings are dramatically reduced.
DWI performance and DWI-based tumour characterisation: DWI offers complementary value by measuring the Brownian motion of water molecules in tissue. Restricted diffusion indicates hypercellularity, which is a feature of viable tumour. For distinguishing tumour recurrence from post-operative inflammation and fibrosis, DWI achieves a specificity of 97%, though its sensitivity is only 60%, meaning it should not be used alone for surveillance but rather as a complement to DCE-MRI and conventional sequences. DWI also has implications for surgical planning: a poorly defined tumour margin on DWI indicates true infiltrative growth rather than reactive peritumoral oedema, potentially altering limb-salvage resection margins.
Multiple DWI models for grading: Beyond simple apparent diffusion coefficient (ADC) calculation, researchers have explored multiple DWI quantification models to better characterise tumour microstructural heterogeneity, including the intravoxel incoherent motion (IVIM) model, the stretched-exponential model, and the diffusion kurtosis model. Manikis and colleagues demonstrated that composite parametric maps generated after selecting the optimal model for each tissue region are highly discriminatory in distinguishing low-grade from high-grade STS, a distinction with major implications for treatment planning and prognosis. Magnetic resonance elastography (MRE), which quantifies tissue stiffness and perfusion, has also shown early feasibility for evaluating STS response to radiation therapy.
Positron emission tomography using fluorodeoxyglucose (FDG-PET) measures glucose uptake as a surrogate for tumour metabolic activity. In STS, 98% of adult tumours are FDG-avid, confirming broad applicability of the technique. FDG-PET can predict STS histological grade at primary diagnosis and is particularly sensitive for high-grade lesions. Two published meta-analyses have demonstrated that PET/CT is superior to both MRI and CT alone for the detection of nodal metastases and soft tissue metastases. Despite these advantages, routine FDG-PET/CT in the initial staging workup of STS has been shown to alter management in fewer than 5% of cases, which has limited its adoption as a standard staging investigation.
Limitations in surveillance: The major limitation of FDG-PET/CT in post-treatment surveillance is the overlap between STS standardised uptake values and those of post-surgical or post-radiation inflammatory change, which can persist for years following treatment. The post-operative resection bed therefore frequently yields non-specific PET findings, reducing diagnostic confidence in the setting where MRI recurrence detection is most important. PET/CT serves best as a problem-solving adjunct: when MRI is contraindicated (e.g., due to metallic implants), when MRI findings are equivocal, or when identifying the metabolically active periphery of a tumour to target biopsy after a central non-diagnostic CT-guided sample.
Hybrid PET/MRI: PET/MRI combines the anatomical, functional, and metabolic information of both modalities simultaneously, representing a potential step beyond PET/CT. One published study demonstrated both a higher detection rate for STS recurrence with PET/MRI compared to MRI alone and greater diagnostic confidence. The technique is particularly promising for reducing ionising radiation compared to PET/CT, which is meaningful for young STS patients who may require many surveillance studies over decades. However, hybrid PET/MRI in STS is still in its early development phase, with limited published evidence on its role in staging, treatment response assessment, and radiation dose optimisation. Standardisation of acquisition protocols and wider scanner availability are prerequisites for broader clinical adoption.
The review identifies artificial intelligence as an emerging contributor to STS imaging, though it acknowledges that the evidence base is still limited. The most developed application described concerns predicting tumour response to radiotherapy using longitudinal DWI combined with deep learning. The rationale is that DWI can detect early treatment-induced changes in tumour cellularity before macroscopic anatomical changes become visible on conventional MRI, providing a biologically meaningful early signal for treatment efficacy. If this signal can be reliably interpreted by a machine learning model, it could enable adaptive radiotherapy planning, where treatment fields are modified mid-course based on early response data.
Deep neural network with GAN-based augmentation: The study cited trained a deep neural network to predict STS pathological response to radiotherapy from longitudinal DWI data. A generative adversarial network (GAN) was used for data augmentation, a technique that creates synthetic training examples to expand a small dataset, addressing the fundamental limitation of small STS cohort sizes. The trained prediction network reported accuracies of 97.1% for patient-based prediction and 83.3% for slice-based prediction on independent test patients. These are promising preliminary figures, but the study is explicitly characterised as limited in scope, with no detail given in the review regarding cohort size, validation methodology, or generalisability across STS subtypes.
DWI quantification models for grading: A complementary AI-adjacent application involves the design of composite parametric maps from multiple DWI quantification models. Rather than relying on a single ADC value, this approach applies the IVIM model, stretched-exponential model, and diffusion kurtosis model in parallel, then selects the optimal model for each region of interest before constructing a composite map. Manikis and colleagues found this approach to be significantly and highly discriminatory in separating low-grade from high-grade STS, which has direct implications for prognosis, chemotherapy decisions, and surgical planning. This represents a radiomics-adjacent methodology even if not explicitly labelled as such.
Context and caution: The review contextualises AI contributions as likely to serve as "adjuncts and problem-solving tools" rather than replacements for standard imaging protocols. The authors note these techniques are expected to be performed predominantly in specialist sarcoma centres due to the advanced training required for image acquisition, processing, and interpretation. The small dataset problem is a recurring obstacle across AI in rare cancers: STS comprises more than 50 histological subtypes, making subtype-specific model training particularly data-hungry and challenging to validate in prospective, multi-centre cohorts.
The review identifies several unresolved standardisation and evidence gaps across STS imaging. For FDG-PET/CT, there is currently no standardised clinical role in STS diagnosis or staging, and the authors call for further research assessing its efficacy in initial staging and re-staging. The overlap of STS standardised uptake values with post-treatment inflammatory changes remains a fundamental limitation that reduces its utility in the surveillance setting where it might otherwise be most valuable. The authors specifically note that FDG-PET has proved limited for detecting metastatic myxoid liposarcoma, pointing clinicians toward WBMRI in that specific context.
Gadolinium retention concerns: A notable limitation surrounding contrast-enhanced MRI is mounting evidence of gadolinium retention in bone and brain following repeated contrast administrations. The long-term implications of this deposition are currently unknown, but regulatory bodies including the Medicines and Healthcare Products Regulatory Agency (MHRA) and European Medicines Agency have revised guidance on gadolinium use. This creates clinical pressure to reduce unnecessary contrast administrations, reinforcing the importance of non-contrast protocols (including DWI and WBMRI) where sufficient diagnostic information can be obtained without gadolinium.
Hybrid PET/MRI development: Hybrid PET/MRI is described as being in its infancy for STS imaging. Key research priorities include defining its role in improving staging accuracy, evaluating treatment response beyond what individual modalities can achieve, and quantifying its advantage in reducing ionising radiation exposure compared to serial PET/CT. Scanner availability and the complexity of simultaneous data acquisition remain practical barriers to widespread clinical adoption.
AI and the rarity problem: For artificial intelligence applications, the rarity of STS is both a scientific challenge and an operational one. With more than 50 histological subtypes, even at a specialist centre the annual case volume for individual subtypes can be very small, making supervised machine learning model training and prospective validation difficult. The GAN-based augmentation approach described represents one strategy for addressing small dataset size, but external validation across multiple institutions and subtypes remains the necessary next step before any AI tool could enter clinical practice. Future progress will likely depend on international multi-centre dataset pooling, standardised DWI acquisition protocols, and prospective study designs.