Large vessel occlusion detection by non-contrast CT using artificial ıntelligence

INTRODUCTION

Computer vision models have been used to diagnose some disorders using computer tomography (CT) and magnetic resonance (MR) images. In this work, our objective is to detect large and small brain vessel occlusion using a deep feature engineering model in acute of ischemic stroke.

METHODS

We use our dataset. which contains 324 patient's CT images with two classes; these classes are large and small brain vessel occlusion. We divided the collected image into horizontal and vertical patches. Then, pretrained AlexNet was utilized to extract deep features. Here, fc6 and fc7 (sixth and seventh fully connected layers) layers have been used to extract deep features from the created patches. The generated features from patches have been concatenated/merged to generate the final feature vector. In order to select the best combination from the generated final feature vector, an iterative selector (iterative neighborhood component analysis-INCA) has been used, and this selector has chosen 43 features. These 43 features have been used for classification. In the last phase, we used a kNN classifier with tenfold cross-validation.

RESULTS

By using 43 features and a kNN classifier, our AlexNet-based deep feature engineering model surprisingly attained 100% classification accuracy.

CONCLUSION

The obtained perfect classification performance clearly demonstrated that our proposal could separate large and small brain vessel occlusion detection in non-contrast CT images. In this aspect, this model can assist neurology experts with the early recanalization chance.

Spine Cryoablation: A Multimodality Image-Guided Approach for Tumors Adjacent to Major Neural Elements

We report percutaneous cryoablation of spine tumors in 7 consecutive patients (5 men, 2 women [mean age, 47 years; range, 17-68 years]) by using intraprocedural image monitoring of ice ball margins to protect adjacent neural elements. Complete tumor ablation was achieved in all patients without neurologic complication. Pain relief was achieved in 4 of 5 (80%) patients; the patient with persistent pain was later found to have enlarging metastases at other sites.

PET/CT-guided percutaneous liver mass biopsies and ablations: targeting accuracy of a single 20 s breath-hold PET acquisition

AIM

To determine whether a single 20 s breath-hold positron-emission tomography (PET) acquisition obtained during combined PET/computed tomography (CT)-guided percutaneous liver biopsy or ablation procedures has the potential to target 2-[(18)F]-fluoro-2-deoxy-d-glucose (FDG)-avid liver masses as accurately as up to 180 s breath-hold PET acquisitions.

MATERIALS AND METHODS

This retrospective study included 10 adult patients with 13 liver masses who underwent FDG PET/CT-guided percutaneous biopsies (n = 5) or ablations (n = 5). PET was acquired as nine sequential 20 s, monitored, same-level breath-hold frames and CT was acquired in one monitored breath-hold. Twenty, 40, 60, and 180 s PET datasets were reconstructed. Two blinded readers marked tumour centres on randomized PET and CT datasets. Three-dimensional spatial localization differences between PET datasets and either 180 s PET or CT were analysed using multiple regression analyses. Statistical tests were two-sided and p < 0.05 was considered significant.

RESULTS

Targeting differences between 20 s PET and 180 s PET ranged from 0.7-20.3 mm (mean 5.3 ± 4.4 mm; median 4.3) and were not statistically different from 40 or 60 s PET (p = 0.74 and 0.91, respectively). Targeting differences between 20 s PET and CT ranged from 1.4-36 mm (mean 9.6 ± 7.1 mm; median 8.2 mm) and were not statistically different from 40, 60, or 180 s PET (p = 0.84, 0.77, and 0.35, respectively).

CONCLUSION

Single 20 s breath-hold PET acquisitions from PET/CT-guided percutaneous liver procedures have the potential to target FDG-avid liver masses with equivalent accuracy to 180 s summed, breath-hold PET acquisitions and may facilitate strategies that improve image registration and shorten procedure times.

Noninvasive multidetector computed tomography enterography in patients with small-bowel Crohn's disease: is a 40-second delay better than 70 seconds?

BACKGROUND

Multidetector computed tomography (MDCT) enterography combines neutral enteric contrast with intravenously administered contrast material. The optimal intravenous (IV) contrast material protocol has still not been established.

PURPOSE

To determine the optimal delay time to image patients with small-bowel Crohn's disease during MDCT enterography.

MATERIAL AND METHODS

After oral administration of 1350 ml of neutral contrast medium, 26 patients with small-bowel Crohn's disease underwent MDCT enterography;scans were obtained 40 s (enteric phase) and 70 s (parenchymal phase) after IV administration of 100 ml of iodinated contrast material. Three radiologists, blinded to clinical and pathological findings, independently and retrospectively evaluated each scan in two separate reading sessions for the presence or absence of CT features of Crohn's disease activity. The interobserver agreement was evaluated, and the efficacy of each phase in detecting active disease in the terminal ileum for each reader was determined. The gold standard was pathology (n=13), endoscopy (n=3), and clinical evaluation (n=10).

RESULTS

No statistically significant difference was present between the enteric and the parenchymal phase for each reader in each segment regarding the presence or absence of CT features of Crohn's disease. The interobserver agreement for the presence of five main features of active Crohn's disease in the terminal ileum ranged from poor to excellent.The sensitivity, specificity, negative predictive value, positive predictive value, and accuracy for active Crohn's disease in the terminal ileum ranged from 40 to 90%, 88 to 100%, 70 to 94%, 44 to 100%, and 69 to 96%, respectively. There was no statistical difference between the two phases for each reader.

CONCLUSION

MDCT enterography in patients with suspected active Crohn's disease can be obtained at either 40 s or 70 s after IV contrast material.

Non-rigid registration of pre-procedural MR images with intra-procedural unenhanced CT images for improved targeting of tumors during liver radiofrequency ablations

In the United States, unenhanced CT is currently the most common imaging modality used to guide percutaneous biopsy and tumor ablation. The majority of liver tumors such as hepatocellular carcinomas are visible on contrast-enhanced CT or MRI obtained prior to the procedure. Yet, these tumors may not be visible or may have poor margin conspicuity on unenhanced CT images acquired during the procedure. Non-rigid registration has been used to align images accurately, even in the presence of organ motion. However, to date, it has not been used clinically for radiofrequency ablation (RFA), since it requires significant computational infrastructure and often these methods are not sufficient robust. We have already introduced a novel finite element based method (FEM) that is demonstrated to achieve good accuracy and robustness for the problem of brain shift in neurosurgery. In this current study, we adapt it to fuse pre-procedural MRI with intra-procedural CT of liver. We also compare its performance with conventional rigid registration and two non-rigid registration methods: b-spline and demons on 13 retrospective datasets from patients that underwent RFA at our institution. FEM non-rigid registration technique was significantly better than rigid (p < 10-5), non-rigid b-spline (p < 10-4) and demons (p < 10-4) registration techniques. The results of our study indicate that this novel technology may be used to optimize placement of RF applicator during CT-guided ablations.