Cancer Imaging最新文献

Background: With the advancements in imaging technologies and the widespread use of magnetic resonance imaging (MRI), the detection rates of pancreatic cystic lesions (PCLs) have significantly increased. While most of these lesions are benign, branch-duct intraductal papillary mucinous neoplasms (BD-IPMNs) pose a potential risk for malignant transformation, necessitating regular clinical and radiological follow-up. However, conventional MRI protocols are time-consuming and resource-intensive, prompting the need for shorter, cost-effective alternatives without compromising diagnostic accuracy. This study aims to evaluate the diagnostic performance and clinical feasibility of abbreviated MRI (A-MRI) protocols for BD-IPMN surveillance compared to standard MRI (S-MRI).

Methods: This was a single-center, retrospective study including patients with BD-IPMN who underwent follow-up MRI between January 2022 and December 2024. Three MRI protocols were analyzed: (1) S-MRI, comprising T2-weighted imaging, dynamic contrast-enhanced (DCE) T1-weighted imaging, 3D MR cholangiopancreatography (MRCP), and diffusion-weighted imaging (DWI); (2) A-MRI protocol 1 (A-MRI-1), including MRCP and T2-weighted sequences; and (3) A-MRI protocol 2 (A-MRI-2), incorporating MRCP, T2-weighted, and DWI sequences. The images were evaluated for lesion size progression (≥ 5 mm in 2 years), mural nodules, cyst wall thickening, main pancreatic duct dilation, and parenchymal atrophy. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for A-MRI protocols in detecting degeneration signs.

Results: A total of 124 patients (mean age: 64 years, 55.6% female) and 124 lesions were analyzed. The detection rates of key degeneration markers were similar between S-MRI and A-MRI protocols, except for contrast-enhanced mural nodules, which were not identifiable with A-MRI due to the lack of DCE sequences. The overall sensitivity, specificity, PPV, and NPV for A-MRI in detecting BD-IPMN degeneration markers were 100%, 98.3%, 71.4%, and 100%, respectively. A-MRI protocols demonstrated a comparable diagnostic performance to S-MRI while significantly reducing scan time (from ~ 40-50 min to 7-12 min). However, false-positive mural nodule detection was higher with A-MRI, potentially leading to unnecessary follow-up imaging. The addition of DWI in A-MRI-2 did not provide a significant diagnostic advantage over A-MRI-1.

Conclusions: A-MRI is a viable alternative for BD-IPMN follow-up, offering substantial reductions in imaging duration and costs while maintaining high diagnostic accuracy. However, the absence of DCE sequences may lead to false-positive mural nodule detection, necessitating further evaluation in selected cases.

Background: New factors predicting response in patients with a PD-L1 tumor proportion score (TPS) ≥ 50% for locally advanced or metastatic non-small cell lung cancer (NSCLC) are needed to better select first-line therapy. Based on the literature, we previously developed a radiomic model predicting the KEAP1/NFE2L2 mutational status.

Method: This was a retrospective monocenter study including 94 consecutive patients with advanced or metastatic PD-L1 ≥ 50% NSCLC, treated with pembrolizumab, who underwent a pre-therapeutic FDG-PET/CT and were followed up for 1 year. Seventy-seven patients who did not progress within the first 60 days of treatment were analyzed. Each primary lesion was segmented by 2 physicians on PET and CT scans. Radiomic features were calculated using MIM software on both PET and CT imaging. A previously developed KEAP1/NFE2L2 radiomic prediction model (called MUT_PET) was applied to this cohort using the initial FDG-PET/CT. The primary endpoint was the validation of the MUT_PET model as a predictive factor of PFS via the non-invasive prediction of KEAP1/NEF2L2 mutation.

Results: The main characteristics of this cohort were: median age of 67.0 years [range, 48.0-84.0], sex ratio M/F = 60/17, 74.0% of patients with a histopathology of adenocarcinoma and 85.0% with a stage IV disease. The median follow-up was 20.0 months [range, 15.3-23.9]. Fifty-six (72.2%) patients experienced a disease progression with a median PFS of 11.8 months (CI95% 8.6-15.8) among which 51 (66.2%) died. In univariable analysis, MUT_PET model was statistically significant as a predictive factor of improved PFS (HR = 0.51, CI95% 0.30-0.91, p = 0.02) whereas it was not statistically significant regarding OS (HR = 0.61, CI95% 0.34-1.11, p = 0.10). In multivariable analysis, the MUT_PET model was associated with a HR of 0.6 (CI95% 0.34-1.06, p = 0.08). Combining the MUT_PET prediction, the histology subtype and the existence of liver metastases, a multimodal nomogram was able to predict PFS (chi-test of 39.43, p < 0.0001).

Conclusion: In PD-L1 TPS ≥ 50% NSCLC patients treated with pembrolizumab, our results suggest an improved PFS in patients predicted to be KEAP1/NFE2L2 mutated.

Objectives: CD19-specific chimeric antigen receptor T-cell therapy (CART) has emerged as effective treatment for relapsed or refractory (r/r) lymphoma. The maximum distance (Dmax) of lymphoma lesions holds potential as prognostic imaging biomarker in lymphoma treated with conventional therapies, but evidence in the context of CART remains scarce and further studies are needed to clarify its clinical relevance. We evaluated Dmax at baseline imaging as a potential prognostic tool for assessment of metabolic and overall response, progression-free survival (PFS) and overall survival (OS).

Material & methods: Consecutive r/r lymphoma patients with (PET/)CT imaging at baseline (BL) before lymphodepletion and subsequent CAR T-cell transfusion were included. Dmax was measured in cm at BL. Patients were divided by tertiles into three equal sized groups according to Dmax. Ann Arbor stages were calculated at baseline and the sum of product diameters (SPD) was used to represent tumor burden (TB). Overall response according to Lugano criteria and the Deauville score were determined at day 90 PET/CT imaging.

Results: Thirty-nine patients met the inclusion criteria. Median Dmax was 40.0 cm (IQR: 16.4-70.3 cm) at BL. Median TB decreased from BL with 4,095 mm² to 770 mm² at FU imaging. Median TB at BL was significantly higher in the Dmax intermediate and high group compared to the Dmax low group (p = 0.005) with 7,222 mm² (IQR: 3,355-11,941 mm²), 4,649 mm² (IQR: 2,376-10,406 mm²) and 1,739 mm² (IQR: 715-7,402 mm²), respectively. Dmax intermediate and high group showed significantly higher Ann Arbor stages (p < 0.001). The survival analysis revealed a significantly (p = 0.030) shorter PFS in the Dmax high group compared to the other patients (91 vs. 364 days), but no relevant differences in OS (p = 0.151).

Conclusions: Patients with high Dmax showed a shorter PFS, but no significant differences in OS. Dmax is a useful parameter for characterizing tumor dissemination, which could also be incorporated into scores due to its interval scale.

Objectives: Lung squamous cell carcinoma (LSCC) is a prevalent subtype of lung cancer characterized by a high mortality rate. Early diagnosis and preoperative prognostic prediction are critical for effective patient stratification, guiding clinical management, and ultimately improving patient outcomes. This study aims to investigate the imaging appearances, dynamic changes during follow-up CT, and surgical prognosis of stage I LSCC.

Methods: A retrospective analysis was performed on 363 patients with central type (c-LSCC) and 310 patients with peripheral type (p-LSCC). The imaging features of c-LSCC and p-LSCC were first established. Subsequently, the dynamic changes of both types during follow-up CT scans were assessed and compared. Finally, the surgical prognosis for each LSCC type were evaluated.

Results: The c-LSCC was categorized into two types: type I (endobronchial lesions; 48.76%, 177/363) and type II (hilar nodule or mass; 51.24%, 186/363). Four CT signs were identified: localized bronchial wall thickening (21.49%, 78/363), soft tissue in the bronchial lumen (9.09%, 33/363), bronchial cast shadows (BCS) (76.03%, 276/363), and hilar nodule or mass (51.24%, 186/363). By comparison, the p-LSCC was classified into three categories: type I (solid nodule; 93.22%, 289/310), type II (thin-walled cystic nodule; 6.13%, 19/310), and type III (subsolid nodule; 0.65%, 2/310). Notably, c-LSCC and p-LSCC showed distinct CT evolution patterns. The c-LSCC advanced from endobronchial growth to airway obstruction and BCS development, ultimately forming a hilar mass. While the p-LSCC showed progressive enlargement with malignant features. Stage I c-LSCC had worse disease-free survival (DFS) and overall survival (OS) than p-LSCC (all p < 0.05). Survival analysis revealed that a CT score of BCS ≥ 4 and obstructive pneumonia were independent predictors of poor outcomes for both DFS and OS in patients with c-LSCC (p < 0.05). Conversely, a history of previous malignancy and perifocal fibrosis predicted unfavorable prognosis for both DFS and OS in patients with p-LSCC (p < 0.05).

Conclusion: Early-stage LSCC demonstrates distinct imaging characteristics, CT evolution patterns, and surgical prognosis between central and peripheral subtypes, informing early diagnosis and personalized management strategies.

Background: Brain tumor classification using Magnetic Resonance Imaging (MRI) is crucial for diagnosis and treatment planning. The differentiation between malignant and benign brain tumors and their subtypes remains a challenging task that can benefit from advanced computational techniques.

Purpose: This study uses an MRI dataset to explore the effectiveness of deep learning (DL) and machine learning (ML) approaches for classifying brain tumors.

Materials and methods: A dataset comprising 1200 DICOM brain tumor MRI images, representing malignant and benign tumors with six subtypes, was prepared. Each image was converted to a 512 × 512-pixel digital format, selecting 200 images per tumor class. Image quality was enhanced using sharpening algorithms and mean filtering. The proposed edge refined binary histogram segmentation (ER-BHS) was applied to extract hybrid features from the regions of interest. Feature optimization through a correlation-based method reduced the dataset to 11 key features. Multiple classifiers, including DL, neural networks, and ML models, were evaluated on the optimized dataset using 10-fold cross-validation.

Results: Among the tested models, the random committee (RC) classifier demonstrated superior performance, achieving an accuracy of 98.61% on the optimized hybrid brain tumor MRI dataset. Overall, DL and ML methods effectively automated brain tumor classification.

Conclusion: The promising results affirm the potential of DL and ML approaches to enhance medical image analysis and improve diagnostic accuracy in brain tumor classification, potentially revolutionizing clinical workflows.

Ovarian cancer (OC) is a leading cause of gynecologic cancer mortality, poses significant diagnostic challenges due to its inter- and intra-tumoral heterogeneity. Conventional qualitative imaging often fails to capture such complexity, whereas radiomics is specifically designed to address it. By integrating imaging features with clinical, genomic, or proteomic data, these approaches reveal sub-visual heterogeneity and enhance diagnostic accuracy. Ultrasound-based radiomics and radiogenomics further extend its clinical utility. Virtual biopsy techniques, such as radiomic habitat maps fused with ultrasound, enable real-time, multi-site sampling to predict prognosis, peritoneal metastasis, and recurrence. This review synthesizes recent advancements in ultrasound-based radiomics and radiogenomics for OC, focusing on their clinical utility in improving diagnostic accuracy, prognostic stratification, and personalized therapeutic strategies. Despite progress, challenges such as insufficient standardization and limited model interpretability persist. Future efforts should prioritize AI-augmented analytical pipelines, multicenter prospective validation, and biomarker discovery to bridge imaging phenotypes with precision oncology framework in OC management.

Introduction: Bevacizumab, an angiogenesis inhibitor, is commonly used alongside chemotherapy for metastatic colorectal cancer (mCCR). While inducing necrosis in tumours, bevacizumab may also lead to atrophy in tumour-free organs. Artificial intelligence (AI) models offer user-friendly methods for measuring organ volumes. This study explores the relationship between bevacizumab-induced atrophy using AI-assisted volume measurement in tumour-free organs and treatment efficacy.

Methods: This multicenter retrospective study includes patients from the PRODIGE 9 and PRODIGE 20 trials. Organ atrophy was assessed by evaluating volume changes from diagnosis to two months after treatment initiation in patients receiving bevacizumab compared to those who did not. Statistical analyses were performed using the Wilcoxon test, with correlations between volumetric changes. Overall and progression-free survival were assessed using log-rank tests and Cox regression models.

Results: Among the 214 patients included, 192 received bevacizumab. Both liver and spleen volumes were measured using a deep learning-based AI model and manual measurements. AI-generated volume measurements showed a strong correlation with manual measurements (Pearson coefficient > 0.8). Bevacizumab-treated patients exhibited significant atrophy of non-tumoural liver volume (p = 0.0378), while no significant changes were observed in tumour or spleen volumes in either group. Survival analyses revealed that patients with a smaller decrease in non-tumoural liver volume had improved overall survival (p = 0.016), although this association became non-significant after adjusting for age, sex, and tumour volume at diagnosis (p = 0.25).

Conclusion: Our findings support the feasibility and reliability of AI in organ volume measurement. While bevacizumab exposure was linked to non-tumoural liver atrophy, its impact on survival remains inconclusive after adjustment. These results pave the way for further research into bevacizumab-induced organ atrophy and the potential of AI in personalizing oncology treatments.