Improved early outcome prediction by MRI-based 3D tumor volume assessment in patients with CNS lymphomas.

BACKGROUND: Central nervous system lymphomas (CNSL) display remarkable clinical heterogeneity, yet accurate prediction of outcomes remains challenging. The IPCG criteria are widely used in routine practice for the assessment of treatment response. However, the value of the IPCG criteria for ultimate outcome prediction is largely unclear, mainly due to the uncertainty in delineating complete from partial responses during and after treatment. METHODS: We explored various MRI features including semi-automated 3D tumor volume measurements at different disease milestones and their association with survival in 93 CNSL patients undergoing curative-intent treatment. RESULTS: At diagnosis, patients with more than 3 lymphoma lesions, periventricular involvement, and high 3D tumor volumes showed significantly unfavorable PFS and OS. At first interim MRI during treatment, the IPCG criteria failed to discriminate outcomes in responding patients. Therefore, we randomized these patients into training and validation cohorts to investigate whether 3D tumor volumetry could improve outcome prediction. We identified a 3D tumor volume reduction of ≥97% as the optimal threshold for risk stratification (=3D early response, 3D_ER). Applied to the validation cohort, patients achieving 3D_ER had significantly superior outcomes. In multivariate analyses, 3D_ER was independently prognostic of PFS and OS. Finally, we leveraged prognostic information from 3D MRI features and circulating biomarkers to build a composite metric that further improved outcome prediction in CNSL. CONCLUSIONS: We developed semi-automated 3D tumor volume measurements as strong and independent early predictors of clinical outcomes in CNSL patients. These radiologic features could help improve risk stratification and help guide future treatment approaches.

Mesoscopic Assessment of Microstructure in Glioblastomas and Metastases by Merging Advanced Diffusion Imaging with Immunohistopathology.

BACKGROUND AND PURPOSE: Glioblastomas and metastases are the most common malignant intra-axial brain tumors in adults and can be difficult to distinguish on conventional MR imaging due to similar imaging features. We used advanced diffusion techniques and structural histopathology to distinguish these tumor entities on the basis of microstructural axonal and fibrillar signatures in the contrast-enhancing tumor component. MATERIALS AND METHODS: Contrast-enhancing tumor components were analyzed in 22 glioblastomas and 21 brain metastases on 3T MR imaging using DTI-fractional anisotropy, neurite orientation dispersion and density imaging-orientation dispersion, and diffusion microstructural imaging-micro-fractional anisotropy. Available histopathologic specimens (10 glioblastomas and 9 metastases) were assessed for the presence of axonal structures and scored using 4-level scales for Bielschowsky staining (0: no axonal structures, 1: minimal axonal fragments preserved, 2: decreased axonal density, 3: no axonal loss) and glial fibrillary acid protein expression (0: no glial fibrillary acid protein positivity, 1: limited expression, 2: equivalent to surrounding parenchyma, 3: increased expression). RESULTS: When we compared glioblastomas and metastases, fractional anisotropy was significantly increased and orientation dispersion was decreased in glioblastomas (each P < .001), with a significant shift toward increased glial fibrillary acid protein and Bielschowsky scores. Positive associations of fractional anisotropy and negative associations of orientation dispersion with glial fibrillary acid protein and Bielschowsky scores were revealed, whereas no association between micro-fractional anisotropy with glial fibrillary acid protein and Bielschowsky scores was detected. Receiver operating characteristic curves revealed high predictive values of both fractional anisotropy (area under the curve = 0.8463) and orientation dispersion (area under the curve = 0.8398) regarding the presence of a glioblastoma. CONCLUSIONS: Diffusion imaging fractional anisotropy and orientation dispersion metrics correlated with histopathologic markers of directionality and may serve as imaging biomarkers in contrast-enhancing tumor components.

A Context-based Chatbot Surpasses Trained Radiologists and Generic ChatGPT in Following the ACR Appropriateness Guidelines.

Background Radiological imaging guidelines are crucial for accurate diagnosis and optimal patient care as they result in standardized decisions and thus reduce inappropriate imaging studies. Purpose In the present study, we investigated the potential to support clinical decision-making using an interactive chatbot designed to provide personalized imaging recommendations from American College of Radiology (ACR) appropriateness criteria documents using semantic similarity processing. Methods We utilized 209 ACR appropriateness criteria documents as specialized knowledge base and employed LlamaIndex, a framework that allows to connect large language models with external data, and the ChatGPT 3.5-Turbo to create an appropriateness criteria contexted chatbot (accGPT). Fifty clinical case files were used to compare the accGPT's performance against general radiologists at varying experience levels and to generic ChatGPT 3.5 and 4.0. Results All chatbots reached at least human performance level. For the 50 case files, the accGPT performed best in providing correct recommendations that were "usually appropriate" according to the ACR criteria and also did provide the highest proportion of consistently correct answers in comparison with generic chatbots and radiologists. Further, the chatbots provided substantial time and cost savings, with an average decision time of 5 minutes and a cost of 0.19 € for all cases, compared to 50 minutes and 29.99 € for radiologists (both p < 0.01). Conclusion ChatGPT-based algorithms have the potential to substantially improve the decision-making for clinical imaging studies in accordance with ACR guidelines. Specifically, a context-based algorithm performed superior to its generic counterpart, demonstrating the value of tailoring AI solutions to specific healthcare applications.