Radiology. Imaging cancer最新文献

Purpose To compare diffusion-weighted (DWI) MRI and PET/CT for diagnosing local-regional residual disease after curative-intent chemoradiation therapy (CRT) in oropharyngeal squamous cell carcinoma (OPSCC), including evaluation of DWI for clarifying equivocal PET/CT findings. Materials and Methods In this prospective study, consecutive participants with OPSCC treated with curative-intent CRT were enrolled between October 2018 and September 2021. DWI was added to the routine PET/CT protocol 3-3.5 months after treatment for local-regional residual disease detection. Reference standards were histopathologic confirmation or unequivocal progression or resolution at follow-up imaging. During qualitative evaluation, imaging findings were classified as negative, equivocal, or positive for residual disease; equivocal findings were considered positive for analysis. Sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were calculated, with differences between modalities assessed using the McNemar test. As a secondary analysis, a sequential imaging strategy using PET/CT and DWI was evaluated. Results A total of 95 participants (mean ± SD age, 61.3 years ± 9.3; 72 male, 85 p16-positive) were included, of whom eight (8.4%) had local-regional residual disease. Sensitivity and negative predictive value for local-regional residual disease detection were 100% for both PET/CT and DWI (eight of eight and 61 of 61 at PET/CT; eight of eight and 72 of 72 at DWI). DWI demonstrated higher specificity (83% [72 of 87] vs 70% [61 of 87]; P < .05) and positive predictive value (35% [eight of 23] vs 24% [eight of 34]; P < .05) than PET/CT. In the sequential imaging analysis, DWI resolved 14 of 34 positive or equivocal PET/CT findings, whereas PET/CT was negative in three of 23 positive or equivocal DWI cases. Conclusion Both PET/CT and DWI demonstrated excellent sensitivity for detecting local-regional residual disease after CRT in OPSCC, as no residual primary tumors or nodal metastases were missed by either modality. DWI showed higher specificity and positive predictive value than PET/CT and demonstrated potential to clarify equivocal PET/CT findings. Keywords: PET/CT, MR-Functional Imaging, MR-Diffusion Weighted Imaging, Head/Neck, Comparative Studies Supplemental material is available for this article. © RSNA, 2026.

Purpose To develop and validate a contrast-enhanced CT-based prediction model for identifying occult lymph node metastasis (OLNM) in patients with early-stage non-small cell lung cancer (NSCLC), with the goal of supporting individualized lymph node dissection (LND) strategies. Materials and Methods This retrospective study included patients with preoperative clinical stage I-IIA (cT1-T2bN0M0) solid NSCLC who underwent lobectomy with systematic LND between January 2021 and April 2024. Univariable and multivariable logistic regression analyses were used to identify independent preoperative CT predictors of OLNM and to construct a nomogram. Model performance was assessed using the area under the receiver operating characteristic curve, and specificity was evaluated at a fixed sensitivity of 95%. Results Among 329 patients with solid NSCLC (median age, 65 years; IQR, 58-70 years; 168 male patients), 22.2% (73 of 329) had OLNM, including 47.9% (35 of 73) with N1 and 52.1% (38 of 73) with N2 metastases. Independent predictors of OLNM were a decreased inner margin ratio (odds ratio [OR], 0.02; 95% CI: 0.00, 0.10; P < .001), presence of the lollipop sign (OR, 3.48; 95% CI: 1.87, 6.49; P < .001), and tumor-pleura relationship type II (OR, 6.95; 95% CI: 2.62, 18.44; P < .001) and type III (OR, 13.27; 95% CI: 5.11, 34.45; P < .001). The nomogram achieved an area under the receiver operating characteristic curve of 0.81 (95% CI: 0.76, 0.87), with a sensitivity of 78.1% and specificity of 73.4%; specificity was 39.1% at 95% sensitivity. Conclusion A contrast-enhanced CT-based nomogram incorporating inner margin ratio, lollipop sign, and tumor-pleura relationship enabled effective preoperative risk stratification for OLNM in early-stage solid NSCLC and may aid in tailoring LND strategies. Keywords: Imaging Modality, Lung, Neoplasms-Primary, Thorax Supplemental material is available for this article. © The Author(s) 2026. Published by the Radiological Society of North America under a CC BY 4.0 license.

Purpose To evaluate the performance of a deep learning algorithm (DLA) for detecting liver metastases (LM) in patients with colorectal cancer (CRC) across diverse clinical contexts and compare its accuracy with that of radiologists. Materials and Methods This retrospective, bicentric study included patients with CRC who underwent contrast-enhanced abdominal CT between January 2019 and December 2021. The DLA accuracy was assessed at the per-nodule and per-patient levels and compared with that of a senior (R1) and an in-training (R2) radiologist blinded to each other's results. The LM detection and false detection rates and interreader agreement were determined. Results Among 181 patients with CRC (mean age, 64 years ± 13 [SD]; 102 male), 95 had LM and 86 had no LM. In the per-nodule analysis, the DLA LM detection rate was 81% (227 of 280; 95% CI: 76.1, 85.2), with no difference compared with R1 (79%; 222 of 280; 95% CI: 74.2, 83.6; P = .49) or R2 (76%; 214 of 280; 95% CI: 71.1, 81.0; P = .19). Detection rates of DLA increased with lesion size: less than 10 mm, 55% (51 of 93; 95% CI: 44.7, 64.6); 10-19 mm, 91% (96 of 106; 95% CI: 83.5, 94.8); and 20 mm or more, 99% (80 of 81; 95% CI: 93.3, 99.8). Detection of subcapsular LM was comparable across readers (DLA, 90% [113 of 125; 95% CI: 84.0, 94.4]; R1, 91% [114 of 125; 95% CI: 84.9, 95.0]; R2, 89% [111 of 125; 95% CI: 82.1, 93.2]). False detection rates were low (DLA, 22% [39 of 181; 95% CI: 16.2, 28.1]; R1, 20% [37 of 181; 95% CI: 15.2, 26.9]; R2, 26% [47 of 181; 95% CI: 20.1, 32.8]; DLA vs R1, P = .004; DLA vs R2, P = .01). DLA false positives were mainly biliary dilatations (n = 14) and diaphragmatic indentations (n = 12). Interreader agreement was moderate (κ = 0.63-0.75). Conclusion DLA performance did not differ from radiologists in detecting LM, with consistent results across lesion sizes and locations. Keywords: Imaging Modality, Abdomen, Gastrointestinal, Liver, Oncology, Comparative Studies, Segmentation, Diagnosis, Deep Learning Supplemental material is available for this article. © RSNA, 2026.

Purpose To evaluate the performance of deep learning models integrating multimodal data for predicting microvascular invasion (MVI) in hepatocellular carcinoma and to investigate the impact of different manual annotation methods on performance. Materials and Methods Patients with hepatocellular carcinoma from three institutions were included in this retrospective study; postoperative histopathology served as the reference standard for MVI. Patients from center A were divided into training and internal test sets; patients from centers B and C formed the external test set. Two manual annotations (voxel-level masks, bounding boxes) were performed on MRI scans. Deep learning models were developed using multimodal data. Model performance was evaluated using the receiver operating characteristic, calibration, and decision curve analysis, with area under the receiver operating characteristic curve (AUC) differences tested by the DeLong test. Results A total of 281 patients were included in this study (mean age, 59.05 years ± 11.92 [SD]; 238 male). Single-sequence models achieved internal test AUCs of 0.57-0.76. Multisequence models reached AUCs of 0.86 (95% CI: 0.77, 0.95) with masks and 0.83 (95% CI: 0.73, 0.94) with bounding boxes. Multimodal fusion improved performance (mask: AUC, 0.88 [95% CI: 0.80, 0.96] vs bounding box: AUC, 0.85 [95% CI: 0.75, 0.94]; P = .50), with external test AUCs of 0.77 (95% CI: 0.66, 0.89) and 0.76 (95% CI: 0.64, 0.88), respectively (P = .40). Bounding box reduced time by 53% (mask = 3.24 minutes; bounding box = 1.52 minutes; P < .001). Conclusion Multimodal fusion models improved predictive performance for MVI. Bounding box annotation achieved statistically comparable overall AUC to that of voxel-level masks while improving annotation efficiency. Keywords: Hepatocellular Carcinoma, Microvascular Invasion, MRI, Deep Learning, Annotation Efficiency, Model Visualization Supplemental material is available for this article. © The Author(s) 2026. Published by the Radiological Society of North America under a CC BY 4.0 license.

Purpose To develop L5 peptide-modified gas vesicles (L5-GVs) for US molecular imaging (USMI) of glypican-3 (GPC3) expression in hepatocellular carcinoma (HCC). Materials and Methods This study was conducted from October 2022 to December 2024. L5-GVs were synthesized by conjugating L5 peptides to gas vesicles derived from Halobacterium NRC-1. In vitro binding was evaluated by incubating fluorescein isothiocyanate-labeled L5-GVs or control GVs (con-GVs) with GPC3-positive HepG2 and GPC3-negative A549 cells. In vivo USMI was performed in subcutaneous HepG2 and A549 tumor-bearing BALB/c nude mice (4-6 weeks old; female; 18-22 g) after injection of con-GVs or L5-GVs. The correlation between USMI signal intensity at 10 minutes and tumor GPC3 immunofluorescence staining was calculated. Single-cell suspensions from 10 resected human HCC specimens were incubated with fluorescein isothiocyanate-labeled L5-GVs, and the correlation between L5-GVs adhesion-positive cell rate and immunohistochemical GPC3 expression was calculated. Results L5-GVs (approximately 252.23 nm ± 1.87) produced stronger fluorescence intensity than con-GVs in HepG2 cells (P < .001), whereas no difference was observed in A549 cells (P = .96). L5-GVs generated stronger contrast signal than con-GVs in HepG2 tumor-bearing mice (P = .004), with no difference observed in A549 tumors (P = .82); the signal intensity at 10 minutes after injection correlated with GPC3 expression (R² = 0.89). In patient-derived HCC samples, L5-GVs adhesion-positive cell rate strongly correlated with immunohistochemical GPC3 expression (R² = 0.94). Conclusion GPC3-targeted L5-GVs enabled specific USMI of HCC in preclinical models, with strong correlation to clinical pathology supporting potential translation for early HCC diagnostic imaging. Keywords: Molecular Imaging, Animal Studies, Ultrasound-Contrast, Contrast Agents-Other Supplemental material is available for this article. © RSNA, 2026 See also commentary by Xu in this issue See also editorial by Zhou in this issue.