Synthetic radiographs, including OpenAI’s ChatGPT-generated ones, are not easily distinguishable from authentic radiographs by either radiologists or large language models; physician training and dedicated tools are crucial to mitigate risks.

Recently, there has been tremendous interest on the use of large language models (LLMs) in radiology. LLMs have been employed for various applications in cancer imaging, including improving reporting speed and accuracy via generation of standardized reports, automating the classification and staging of abnormal findings in reports, incorporating appropriate guidelines, and calculating individualized risk scores. Another use of LLMs is their ability to improve patient comprehension of imaging reports with simplification of the medical terms and possible translations to multiple languages. Additional future applications of LLMs include multidisciplinary tumor board standardizations, aiding patient management, and preventing and predicting adverse events (contrast allergies, MRI contraindications) and cancer imaging research. However, limitations such as hallucinations and variable performances could present obstacles to widespread clinical implementation. Herein, we present a review of the current and future applications of LLMs in cancer imaging, as well as pitfalls and limitations.

Purpose: This systematic review and meta-analysis compared the diagnostic performance of MRI, [¹⁸F]FDG-PET/CT, and [¹⁸F]FDG-PET/MRI in detecting focal bone lesions and bone marrow infiltration in the initial staging of patients with multiple myeloma (MM) who underwent both MRI and [¹⁸F]FDG-PET/CT or [¹⁸F]FDG-PET/MRI studies. Patients and Methods: A systematic literature search was conducted across PubMed, Embase, and Cochrane databases, including studies comparing the performance of MRI (WB-MRI or spine/pelvis MRI), [¹⁸F]FDG-PET/CT, and/or [¹⁸F]FDG-PET/MRI in the same patients for MM initial staging. Pooled sensitivities and concordance between imaging modalities were analyzed using R (package META-R). Heterogeneity and bias were assessed with the QUADAS-C tool. Results: Twenty studies (published between 2007 and 2025) using the international MM diagnostic criteria as a reference standard met the inclusion criteria. Of these, 13 (n=742) compared per-patient sensitivity of [¹⁸F]FDG-PET/CT and MRI, and 4 (n=224) compared MRI and [¹⁸F]FDG-PET/MRI. Pooled sensitivities were 0.807 (95% CI: 0.74–0.86) for [¹⁸F]FDG-PET/CT (with significant heterogeneity) versus 0.914 (95% CI: 0.88–0.94) for MRI (0.906 for spine/pelvis MRI and 0.920 for WB-MRI) (P<0.001 for meta-regression analysis). Using contingency tables, 83% (599/721) of included patients had concordant [¹⁸F]FDG-PET/CT and MRI results, while 14% (101/721) patients had negative [¹⁸F]FDG-PET/CT and positive MRI with significant differences between the 2 techniques for the paired sample analysis (P<0.001). The pooled sensitivity of the 4 studies including [¹⁸F]FDG-PET/MRI was 0.944 (95% CI: 0.88–0.98). Consensus definitions for specificity in MM imaging should be standardized across studies. Conclusions: This systematic review and comparative meta-analysis demonstrates superior sensitivity of WB-MRI compared with [¹⁸F]FDG-PET/CT for initial staging of MM patients. Future international guidelines might prioritize MRI and [¹⁸F]FDG-PET/MRI for staging of MM patients. Registration: PROSPERO CRD42024564937.

The large language models ChatGPT-o1 and Gemini-2.5-Pro, guided with Oncologically Relevant Findings Reporting and Data System (ONCO-RADS) prompts, accurately extracted and classified incidental findings from whole-body imaging reports and outperformed these strategies without ONCO-RADS prompts and the medical named entity recognition model.