Osteosarcoma (OS) is the most common primary malignant bone tumor, with an incidence that follows a bimodal age distribution, peaking in adolescents and again in older adults. In the United States, approximately 400 children and adolescents receive an OS diagnosis each year. Unlike many adult solid tumors, OS most commonly spreads to the lungs rather than lymph nodes or other organs, making pulmonary surveillance a core pillar of disease management from initial staging through long-term follow-up.
Prognostic impact of lung metastases: The presence or absence of lung metastases at diagnosis defines two dramatically different prognostic groups. Patients with localized OS can achieve 5-year survival rates of 60-70%, but this figure falls to just 10-30% in the metastatic setting. This survival gap underscores why early, accurate detection of pulmonary lesions, even small ones, is directly tied to treatment decisions and patient outcomes. Approximately 15-20% of patients already have detectable lung metastases at first evaluation, and the majority of these patients have the lung as their only metastatic site.
Scope of the review: Published in Cells (2021) by Chiesa, Spinnato, Miceli, Facchini, and Santucci from the Istituto Ortopedico Rizzoli in Bologna, Italy, this narrative review surveys the full landscape of radiologic tools for assessing osteosarcoma lung metastases. It covers the clinical roles of plain radiography, standard and low-dose CT, volumetric assessment via volume doubling time (VDT), computer-aided diagnosis (CAD) systems, and PET/CT, offering a comparative analysis of their strengths and limitations in this specific pediatric and young adult population.
The review provides a current-state-of-the-art synthesis rather than a systematic review or meta-analysis, drawing on published clinical studies, ESMO guidelines, and emerging AI-based imaging tools to chart a path toward more accurate and radiation-efficient follow-up strategies.
Understanding which patients are at highest risk for lung metastasis shapes follow-up intensity and imaging frequency. The literature has identified several clinical and pathological features that correlate with pulmonary dissemination. Axial OS (arising in the spine or pelvis rather than the extremities) with a primary tumor diameter greater than 5 cm carries a higher risk of lung metastasis at diagnosis. By contrast, patient age and primary tumor location alone are not independent risk factors for dissemination.
Histologic subtype and molecular markers: A large population study of more than 1,000 patients found that specific OS subtypes, including Paget's disease-associated OS and small cell OS, were linked to greater risk of lung dissemination. At the molecular level, cytoplasmic HER-2 expression and elevated inflammatory ratios, including monocyte ratio greater than 1 and neutrophil-to-lymphocyte ratio (NLR) greater than 1, were also associated with increased metastatic risk. These markers have potential as risk-stratification tools to guide the frequency of imaging surveillance.
Pathologic fractures and chemotherapy response: A 2017 meta-analysis of 14 studies suggested that clinical presentation with pathologic fractures may increase the risk of subsequent metastases, though this finding awaits further confirmatory research. Neoadjuvant chemotherapy response, typically assessed by histologic necrosis rate after resection, is another recognized prognostic marker. Patients with poor histologic response (less than 90% tumor necrosis) are generally considered at higher risk for relapse, warranting more intensive radiologic follow-up.
The interplay of these risk factors is important context for understanding why a one-size-fits-all follow-up protocol may be suboptimal, and why risk-stratified imaging approaches, potentially enhanced by AI-driven predictive models, are an active area of investigation.
Chest computed tomography (CT) is the gold standard for detecting osteosarcoma lung metastases, yet its sensitivity in published studies ranges widely, from 56% to 84%. This variability is partly explained by the high frequency of atypical radiologic presentations that deviate from the classic picture of multiple, peripheral, solid round nodules. Radiologists and oncologists must be aware that OS lung metastases can present in forms that overlap considerably with benign pulmonary disease.
Spectrum of atypical appearances: According to Seo et al., the CT morphology of OS lung metastases is heterogeneous and can include calcification (a finding typically associated with benign granulomas or hamartomas), hemorrhagic halo, cavitation, spontaneous pneumothorax, tumor embolism, endobronchial localization, and dilated vessels within a mass. A retrospective analysis by Ciccarese et al. of 283 resected lesions found that approximately 14% of OS metastases were not nodular at all, presenting instead as striae, consolidations, pleural nodules, plaques, cavitated lesions, or ground-glass opacities. Striae in particular were identified as a potential pitfall, as they may be misread as atelectasis or fibrosis.
Unreliability of morphologic criteria for benign-malignant discrimination: A key clinical challenge is that no single CT criterion reliably distinguishes benign from malignant pulmonary nodules in OS patients. Picci et al. demonstrated in a retrospective study that both malignant and benign nodules can change in number and size during chemotherapy, undermining the assumption that nodule behavior on treatment predicts malignancy. Although calcified nodules and those greater than 5 mm were more often malignant in some series, pediatric-specific data showed that nearly 32% of confirmed OS lung metastases were not solid or rounded with well-defined margins.
Post-metastasectomy surveillance: CT interpretation becomes even more difficult after thoracotomy and pulmonary metastasectomy, when surgical scars and staple lines can mimic recurrent disease. Focal pleural thickening remains the most reliable radiologic sign of recurrence in this postoperative context, though distinguishing it from benign postoperative change still requires clinical correlation and often serial imaging.
Given that OS predominantly affects children and young adults, and that the risk of lung metastases remains elevated for at least 5 years after the completion of chemotherapy, patients typically require a high frequency of chest CT scans over an extended surveillance period. The European Society for Medical Oncology (ESMO), in 2018 guidelines developed with the European Reference Network for Rare Adult Solid Cancers (EURACAN) and the European Reference Network for Pediatric Oncology (PedCan), proposed the following CT follow-up schedule after completion of chemotherapy: approximately every 3 months for the first 2 years, every 6 months for years 3 to 5, every 6-12 months for years 5 to 10, and every 0.5 to 2 years thereafter depending on institutional practice.
CT protocol considerations: Standard chest CT for OS surveillance is performed without intravenous contrast unless there is clinical concern for chest wall, hilar, or mediastinal involvement. Guidelines recommend spiral CT with 3-5 mm reconstruction thickness in a single breath-hold, with maximum intensity projection (MIP) review to improve small nodule detection. Thin-slice imaging improves sensitivity for nodules smaller than 5 mm but at the cost of increased false-positive rates, particularly in adult patients.
CT vs. plain radiography: Although some institutions historically used chest X-ray for OS follow-up due to lower cost and radiation exposure, evidence strongly favors CT. Paioli and colleagues demonstrated that CT-based follow-up was associated with a higher rate of achieving second complete remission and better prognosis at 5 years, attributed to earlier detection of recurrence enabling timely and effective surgical resection. Plain radiography retains a role only in the immediate postoperative period, where it can detect complications such as pneumothorax or pleural effusion following metastasectomy.
The ongoing challenge of subcentimeter nodules: Improved CT technology has increased the detection of nodules smaller than 5 mm, many of which are transient benign findings such as infections or interscanner variability artifacts. The Children's Oncology Group (COG) defines metastatic disease as a single nodule greater than 10 mm or more than 3 nodules greater than 5 mm, a threshold that helps reduce clinically insignificant incidental findings from triggering treatment changes. Cipriano et al. found no significant difference in outcomes between patients with no pulmonary nodules and those with a single nodule smaller than 5 mm.
The intensive surveillance schedule mandated for OS patients, particularly children and adolescents, carries a meaningful cumulative radiation burden. A published calculation found that the average total ionizing radiation dose for a pediatric sarcoma patient undergoing the full workup and follow-up (chest radiographs, chest CTs, PET scans, and bone scans) was approximately 37.1 mGy, a figure roughly equivalent to the lifetime occupational dose received by nuclear power plant workers. Given the young age at OS onset and the longer remaining life expectancy of these patients, this cumulative exposure raises genuine concerns about secondary radiation-induced malignancies.
Iterative reconstruction methods: Significant reductions in CT radiation dose have been achieved through iterative reconstruction (IR) algorithms, which replace the traditional filtered back projection (FBP) method. Commercially available IR systems include ASIR (GE Healthcare), AIDR 3D (Toshiba), iDose4 (Philips), and SAFIRE and IRIS (Siemens). These methods produce images with higher signal-to-noise ratios and enhanced low-contrast lesion visibility while delivering dose reductions of 40-50% compared to FBP. A limitation is that aggressive noise reduction produces an "oversmoothing" appearance that can affect the assessment of subtle imaging findings.
Model-based iterative reconstruction (MBIR) and deep learning image reconstruction (DLIR): The next generation of dose reduction comes from model-based iterative reconstruction, commercialized as "Veo" by GE Healthcare. A study by Kim specifically evaluated MBIR in pediatric patients, demonstrating that ultra-low-dose chest CT could be achieved without degrading image quality, even in the subset of 29 out of 57 patients who had confirmed lung metastases. More recently, deep learning image reconstruction (DLIR), using deep convolutional neural networks trained on paired high-dose and low-dose CT data, has been shown to significantly reduce image noise while maintaining superior image quality compared to iterative reconstruction methods, representing the current frontier in low-dose CT.
The push for further DLIR development and validation in OS-specific populations is motivated not only by the radiation safety imperative but also by the potential for these AI-based reconstruction tools to simultaneously improve nodule conspicuity, which could translate into greater sensitivity for small metastases.
Traditional CT response assessment in oncology uses the RECIST criteria, which rely on measuring the longest diameter of target lesions. While RECIST is practical and reproducible for most solid tumors, volumetric evaluation of pulmonary nodules offers meaningful advantages, particularly for small and irregularly shaped lesions. Semi-automated volumetric measurement captures three-dimensional growth more completely than a single diameter, and volume doubling time (VDT) calculates the rate at which a nodule doubles in volume, providing dynamic information about lesion behavior across serial scans.
VDT in lung cancer and sarcoma: Volumetric assessment and VDT are now strongly recommended for lung cancer screening programs, where they have demonstrated superiority over diameter-based assessment in predicting overall survival. In sarcoma, published data show that patients with shorter VDT for lung metastases have significantly worse sarcoma-specific survival compared to those with longer VDT, indicating that growth velocity is an independent prognostic variable beyond simply the number or size of metastatic lesions.
Application to osteosarcoma: The review notes that while VDT has shown promise in sarcoma broadly, targeted research in OS-specific populations is still needed. The relatively small size of many OS lung metastases, combined with their morphologic heterogeneity (including calcified, cavitated, and non-nodular forms), may complicate reliable semi-automated volume measurement in this population. Standardizing volumetric CT protocols and validating VDT thresholds specifically for OS patients represents a research gap with direct clinical implications for surveillance decision-making.
Integrating VDT measurement into routine CT follow-up, potentially assisted by AI-driven automated segmentation and volume computation, could provide oncologists with quantitative growth kinetics that supplement the qualitative Radiologist's assessment and improve consistency across institutions and time points.
Computer-aided diagnosis (CAD) systems have become standard tools in lung cancer CT screening programs, where they function as a computational second reader, flagging nodules that human radiologists might overlook in the large volume of images generated by each scan. For each OS follow-up CT, a radiologist must review thousands of individual image slices. The sheer volume of data, combined with the subjective nature of human visual pattern recognition and the fatigue inherent in high-volume reading sessions, creates meaningful opportunity for missed lesions. CAD systems, by contrast, apply consistent algorithmic criteria across every image slice without fatigue or inter-reader variability.
Performance in lung cancer screening: In lung cancer screening contexts, published data show that CAD systems can detect up to 70% of pulmonary nodules missed by radiologists on initial review. CAD has also been shown to identify lesions that multiple independent radiologists all missed in the same reading session, suggesting that the algorithm captures complementary detection information not fully captured by human visual inspection. Deep learning-based CAD, using convolutional neural networks trained on large annotated CT datasets, is the current state of the art in nodule detection, with performance increasingly approaching or matching that of experienced thoracic radiologists on benchmark datasets.
Potential for osteosarcoma applications: Despite this evidence base, the review notes that CAD systems specifically validated for detecting OS lung metastases, as distinct from primary lung cancer nodules, are not yet established. This distinction matters because OS metastases more frequently display atypical morphologies (calcified, cavitated, ground-glass) compared to primary lung adenocarcinomas or squamous cell carcinomas, which are the predominant lesion types in existing CAD training datasets. A CAD system trained primarily on lung cancer nodules may underperform when applied to OS metastases with different density profiles and shapes.
Path to implementation: The authors advocate for dedicated research building CAD systems trained and validated on OS-specific CT datasets, which would require collaborative data collection across multiple institutions given the relatively small number of OS cases at any single center. Such a system, once validated, could serve as a reliable second reader for radiologists, reducing missed metastases and enabling earlier surgical referral while also providing volume measurements that support VDT calculation.
18F-fluorodeoxyglucose (FDG) PET/CT has an established role in the evaluation of many adult solid tumors, but its application in osteosarcoma carries specific limitations that restrict its utility for pulmonary metastasis detection. The minimum lesion size detectable by PET/CT is approximately 5-6 mm in diameter, set by the spatial resolution limitations of the modality. Because OS lung metastases frequently present as small nodules, often well below this threshold, PET/CT cannot reliably detect or characterize small pulmonary lesions, making CT alone the preferred study for lung nodule assessment in OS patients.
Where PET/CT does add value: PET/CT contributes more definitively to the evaluation of bone metastases, where it demonstrates higher sensitivity than conventional bone scintigraphy. Several published studies suggest that FDG-PET may also serve as a promising tool for assessing response to neoadjuvant chemotherapy in OS, with metabolic response on interim PET correlated with histologic necrosis rates at resection. A prospective single-center trial specifically evaluated FDG-PET/CT for assessing early response to neoadjuvant chemotherapy in pediatric sarcoma patients, supporting this potential application. In this context, PET/CT plays a complementary role to CT, contributing information about metabolic activity and treatment response rather than lung nodule detection per se.
Comparative summary of imaging modalities: The authors present a summary table comparing the four imaging modalities across five performance dimensions. Standard CT scores highest for small lesion detection (less than 1 cm) and is rated high for treatment response evaluation and prognostic relevance. Low-dose CT retains strong detection capability with reduced radiation exposure. PET/CT excels at treatment response evaluation and prognostic relevance but scores very low for small lesion detection and carries high radiation and cost burdens. Plain X-ray is rated low across most dimensions, retaining value only for post-metastasectomy complication assessment.
Future directions: The review concludes by identifying several priorities for advancing OS lung metastasis surveillance. First, OS-specific CAD systems trained on datasets that capture the full morphologic spectrum of OS metastases, including atypical presentations, need to be developed and validated in multicenter prospective studies. Second, VDT protocols need to be standardized and validated specifically for OS populations. Third, DLIR algorithms should be evaluated in OS-specific cohorts to confirm that their dose-reduction benefits do not come at the expense of small metastasis detectability. Integration of AI-assisted reconstruction, automated nodule detection, and volumetric growth tracking into unified clinical platforms could substantially improve the accuracy, consistency, and radiation efficiency of OS surveillance over the typical decade-long follow-up period.