Machine learning for predicting the risk stratification of 1-5 cm gastric gastrointestinal stromal tumors based on CT.

BACKGROUD: To predict the malignancy of 1-5 cm gastric gastrointestinal stromal tumors (GISTs) by machine learning (ML) on CT images using three models - Logistic Regression (LR), Decision Tree (DT) and Gradient Boosting Decision Tree (GBDT). METHODS: 231 patients from Center 1 were randomly assigned into the training cohort (n = 161) and the internal validation cohort (n = 70) in a 7:3 ratio. The other 78 patients from Center 2 served as the external test cohort. Scikit-learn software was used to build three classifiers. The performance of the three models were evaluated by sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV) and area under the curve (AUC). Diagnostic differences between ML models and radiologists were compared in the external test cohort. Important features of LR and GBDT were analyzed and compared. RESULTS: GBDT outperformed LR and DT with the largest AUC values (0.981 and 0.815) in the training and internal validation cohorts and the greatest accuracy (0.923, 0.833 and 0.844) across all three cohorts. However, LR was found to have the largest AUC value (0.910) in the external test cohort. DT yielded the worst accuracy (0.790 and 0.727) and AUC values (0.803 and 0.700) in both the internal validation cohort and the external test cohort. GBDT and LR performed better than radiologists. Long diameter was demonstrated to be the same and most important CT feature for GBDT and LR. CONCLUSIONS: ML classifiers, especially GBDT and LR with high accuracy and strong robustness, were considered to be promising in risk classification of 1-5 cm gastric GISTs based on CT. Long diameter was found the most important feature for risk stratification.

Risk prediction model establishment with tri-phasic CT image features for differential diagnosis of adrenal pheochromocytomas and lipid-poor adenomas: Grouping method.

Objectives: The purpose of this study was to establish a risk prediction model for differential diagnosis of pheochromocytomas (PCCs) from lipid-poor adenomas (LPAs) using a grouping method based on tri-phasic CT image features. Methods: In this retrospective study, we enrolled patients that were assigned to a training set (136 PCCs and 183 LPAs) from two medical centers, along with an external independent validation set (30 PCCs and 54 LPAs) from another center. According to the attenuation values in unenhanced CT (CTu), the lesions were divided into three groups: group 1, 10 HU < CTu ≤ 25 HU; group 2, 25 HU < CTu ≤ 40 HU; and group 3, CTu > 40 HU. Quantitative and qualitative CT imaging features were calculated and evaluated. Univariate, ROC, and binary logistic regression analyses were applied to compare these features. Results: Cystic degeneration, CTu, and the peak value of enhancement in the arterial and venous phase (DEpeak) were independent risk factors for differential diagnosis of adrenal PCCs from LPAs. In all subjects (groups 1, 2, and 3), the model formula for the differentiation of PCCs was as follows: Y = -7.709 + 3.617*(cystic degeneration) + 0.175*(CTu ≥ 35.55 HU) + 0.068*(DEpeak ≥ 51.35 HU). ROC curves were drawn with an AUC of 0.95 (95% CI: 0.927-0.973) in the training set and 0.91 (95% CI: 0.860-0.929) in the external validation set. Conclusion: A reliable and practical prediction model for differential diagnosis of adrenal PCCs and LPAs was established using a grouping method.

Unusual magnetic resonance imaging findings of brain and leptomeningeal metastasis in lung adenocarcinoma: A case report.

BACKGROUND: Metastatic tumors are the most common malignancies of central nervous system in adults, and the frequent primary lesion is lung cancer. Brain and leptomeningeal metastases are more common in patients with non-small-cell lung cancer harboring epidermal growth factor receptor mutations. However, the coexist of brain metastasis with leptomeningeal metastasis (LM) in isolated gyriform appearance is rare. CASE SUMMARY: We herein presented a case of a 76-year-old male with an established diagnosis as lung adenocarcinoma with gyriform-appeared cerebral parenchymal and leptomeningeal metastases, accompanied by mild peripheral edema and avid contrast enhancement on magnetic resonance imaging. Surgical and pathological examinations confirmed the brain and leptomeningeal metastatic lesions in the left frontal cortex, subcortical white matter and local leptomeninges. CONCLUSION: This case was unique with respect to the imaging findings of focal gyriform appearance, which might be caused by secondary parenchymal brain metastatic tumors invading into the leptomeninges or coexistence with LM. Radiologists should be aware of this uncommon imaging presentation of tumor metastases to the central nervous system.

Differentiation of lipid-poor adenoma from pheochromocytoma on biphasic contrast-enhanced CT.

PURPOSE: To evaluate the diagnostic performance of biphasic contrast-enhanced CT in differentiation of lipid-poor adenomas from pheochromocytomas. METHODS: 129 patients with 132 lipid-poor adenomas and 93 patients with 97 pheochromocytomas confirmed by pathology were included in this retrospective study. Patients underwent unenhanced abdominal CT scan followed by arterial and venous phase. Quantitative and qualitative imaging features were compared between the two groups using univariate analysis. Risk factors for pheochromocytomas were evaluated by multivariate logistic regression analysis and a diagnostic scoring model was established based on odd ratio (OR) of the risk factors. RESULTS: Pheochromocytomas were larger and showed cystic degeneration more frequently compared with lipid-poor adenomas (p < 0.01). No significant difference was found in peak enhancement phase between the two groups (p = 0.348). Attenuation values on unenhanced phase (CTU), arterial phase (CTA), and venous phase (CTV) of pheochromocytomas were significantly higher than that of lipid-poor adenomas while enhancement ratio on arterial and venous phase (ERA, ERV) of pheochromocytomas was significantly lower than that of lipid-poor adenomas (all p < 0.05). Multivariate analysis revealed lesion size > 29 mm (OR: 5.74; 95% CI 2.51-13.16; p < 0.001), CTA > 81 HU (OR: 2.54; 95% CI 1.04-6.17; p = 0.04), CTV > 97 HU (OR: 11.19; 95% CI 3.21-38.97; p < 0.001), ERV ≤ 1.5 (OR: 20.23; 95% CI 6.30-64.87; p < 0.001), and the presence of cystic degeneration (OR: 6.22, 95% CI 1.74-22.25; p = 0.005) were risk factors for pheochromocytomas. The diagnostic scoring model yielded an area under the curve (AUC) of 0.911. CONCLUSIONS: Biphasic contrast-enhanced CT showed good diagnostic performance in differentiation of lipid-poor adenomas from pheochromocytomas.

Computed tomography characteristics of adrenal ganglioneuroma: a retrospective analysis of 30 pathologically-confirmed cases.

OBJECTIVE: To investigate the computed tomography (CT) characteristics of adrenal ganglioneuromas (AGNs) and to determine the ability of CT scanning to distinguish between large (>3 cm) and small (≤3 cm) AGNs. METHODS: This retrospective study searched the electronic medical record system of a hospital between January 2008 and July 2019 in order to identify patients with pathologically-confirmed AGNs that underwent three phases of CT scanning. The CT features were compared between large and small AGNs. RESULTS: A total of 30 patients with pathologically-confirmed AGNs were included in the study. The majority of patients (76.7%; 23 of 30) were asymptomatic and there were nonspecific symptoms in seven patients. The 'pointed peach' sign appeared in more than half of the patients (53.3%; 16 of 30). The CT value of the arterial phase, progressive enhancement, morphology and calcification in the CT images were found to be significantly different between large and small AGNs. Progressive enhancement was more likely to occur in small AGNs. Most large AGNs had irregular shapes, while small AGNs were likely to be round or oval with a smooth border. Calcifications were noted in large AGNs (42.9%; six of 14). CONCLUSION: CT scanning can show many of the key imaging characteristics of AGNs used to distinguish between large and small AGNs.

Presurgical localization and spatial shift of resting state networks in patients with brain metastases.

Brain metastases are the most prevalent cerebral tumors. Resting state networks (RSNs) are involved in multiple perceptual and cognitive functions. Therefore, precisely localizing multiple RSNs may be extremely valuable before surgical resection of metastases, to minimize neurocognitive impairments. Here we aimed to investigate the reliability of independent component analysis (ICA) for localizing multiple RSNs from resting-state functional MRI (rs-fMRI) data in individual patients, and further evaluate lesion-related spatial shifts of the RSNs. Twelve patients with brain metastases and 14 healthy controls were recruited. Using an improved automatic component identification method, we successfully identified seven common RSNs, including: the default mode network (DMN), executive control network (ECN), dorsal attention network (DAN), language network (LN), sensorimotor network (SMN), auditory network (AN) and visual network (VN), in both individual patients and controls. Moreover, the RSNs in the patients showed a visible spatial shift compared to those in the controls, and the spatial shift of some regions was related to the tumor location, which may reflect a complicated functional mechanism - functional disruptions and reorganizations - caused by metastases. Besides, higher cognitive networks (DMN, ECN, DAN and LN) showed significantly larger spatial shifts than perceptual networks (SMN, AN and VN), supporting a functional dichotomy between the two network groups even in pathologic alterations associated with metastases. Overall, our findings provide evidence that ICA is a promising approach for presurgical localization of multiple RSNs from rs-fMRI data in individual patients. More attention should be paid to the spatial shifts of the RSNs before surgical resection.

Benign ependymoma with extensive intracranial and spinal cerebrospinal fluid dissemination: case report and literature review.

Myxopapillary ependymoma (MPE) is a rare variant of ependymoma that is most commonly located in the cauda equina and filum terminale. We present a case of 23-year-old man diagnosed with MPE in the fourth ventricle and sacral canal area with extensive disseminated lesions along the cerebrospinal ventricular system. Additionally, a molecular pathological diagnosis was performed. The patient underwent a craniotomy and a lumbar laminectomy. In the course of 18 months of follow-up, the patient have recovered very well.

PreSurgMapp: a MATLAB Toolbox for Presurgical Mapping of Eloquent Functional Areas Based on Task-Related and Resting-State Functional MRI.

The main goal of brain tumor surgery is to maximize tumor resection while minimizing the risk of irreversible postoperative functional sequelae. Eloquent functional areas should be delineated preoperatively, particularly for patients with tumors near eloquent areas. Functional magnetic resonance imaging (fMRI) is a noninvasive technique that demonstrates great promise for presurgical planning. However, specialized data processing toolkits for presurgical planning remain lacking. Based on several functions in open-source software such as Statistical Parametric Mapping (SPM), Resting-State fMRI Data Analysis Toolkit (REST), Data Processing Assistant for Resting-State fMRI (DPARSF) and Multiple Independent Component Analysis (MICA), here, we introduce an open-source MATLAB toolbox named PreSurgMapp. This toolbox can reveal eloquent areas using comprehensive methods and various complementary fMRI modalities. For example, PreSurgMapp supports both model-based (general linear model, GLM, and seed correlation) and data-driven (independent component analysis, ICA) methods and processes both task-based and resting-state fMRI data. PreSurgMapp is designed for highly automatic and individualized functional mapping with a user-friendly graphical user interface (GUI) for time-saving pipeline processing. For example, sensorimotor and language-related components can be automatically identified without human input interference using an effective, accurate component identification algorithm using discriminability index. All the results generated can be further evaluated and compared by neuro-radiologists or neurosurgeons. This software has substantial value for clinical neuro-radiology and neuro-oncology, including application to patients with low- and high-grade brain tumors and those with epilepsy foci in the dominant language hemisphere who are planning to undergo a temporal lobectomy.