Radiomics-based machine learning of MRI studies can be useful to predict the tumor type of brain metastases. PurposeTo investigate the feasibility of tumor type prediction with MRI radiomic image features of different brain metastases in a multiclass machine learning approach for patients with unknown primary lesion at the time of diagnosis.Materials and methodsThis single-center retrospective analysis included radiomic features of 658 brain metastases from T1-weighted contrast material–enhanced, T1-weighted nonenhanced, and fluid-attenuated inversion recovery (FLAIR) images in 189 patients (101 women, 88 men; mean age, 61 years; age range, 32–85 years). Images were acquired over a 9-year period (from September 2007 through December 2016) with different MRI units, reflecting heterogeneous image data. Included metastases originated from breast cancer (n = 143), small cell lung cancer (n = 151), non–small cell lung cancer (n = 225), gastrointestinal cancer (n = 50), and melanoma (n = 89). A total of 1423 quantitative image features and basic clinical data were evaluated by using random forest machine learning algo...

Objective: This work addresses two key problems of skin lesion classification. The first problem is the effective use of high-resolution images with pretrained standard architectures for image classification. The second problem is the high class imbalance encountered in real-world multi-class datasets. Methods: To use high-resolution images, we propose a novel patch-based attention architecture that provides global context between small, high-resolution patches. We modify three pretrained architectures and study the performance of patch-based attention. To counter class imbalance problems, we compare oversampling, balanced batch sampling, and class-specific loss weighting. Additionally, we propose a novel diagnosis-guided loss weighting method which takes the method used for ground-truth annotation into account. Results: Our patch-based attention mechanism outperforms previous methods and improves the mean sensitivity by 7%. Class balancing significantly improves the mean sensitivity and we show that our diagnosis-guided loss weighting method improves the mean sensitivity by 3% over normal loss balancing. Conclusion: The novel patch-based attention mechanism can be integrated into pretrained architectures and provides global context between local patches while outperforming other patch-based methods. Hence, pretrained architectures can be readily used with high resolution images without downsampling. The new diagnosis-guided loss weighting method outperforms other methods and allows for effective training when facing class imbalance. Significance: The proposed methods improve automatic skin lesion classification. They can be extended to other clinical applications where high-resolution image data and class imbalance are relevant.

Background We determined whether a comprehensive assessment of cerebral collateral blood flow is associated with ischemic lesion edema growth in patients successfully treated by thrombectomy. Methods This was a multicenter retrospective study of ischemic stroke patients who underwent thrombectomy treatment of large vessel occlusions. Collateral status was determined using the cerebral collateral cascade (CCC) model, which comprises three components: arterial collaterals (Tan Scale) and venous outflow profiles (Cortical Vein Opacification Score) on CT angiography, and tissue-level collaterals (hypoperfusion intensity ratio) on CT perfusion. Quantitative ischemic lesion net water uptake (NWU) was used to determine edema growth between admission and follow-up non-contrast head CT (ΔNWU). Three groups were defined: CCC+ (good pial collaterals, tissue-level collaterals, and venous outflow), CCC− (poor pial collaterals, tissue-level collaterals, and venous outflow), and CCCmixed (remainder of patients). Primary outcome was ischemic lesion edema growth (ΔNWU). Multivariable regression models were used to assess the primary and secondary outcomes. Results 538 patients were included. 157 patients had CCC+, 274 patients CCCmixed, and 107 patients CCC− profiles. Multivariable regression analysis showed that compared with patients with CCC+ profiles, CCC− (β 1.99, 95% CI 0.68 to 3.30, P=0.003) and CCC mixed (β 1.65, 95% CI 0.75 to 2.56, P<0.001) profiles were associated with greater ischemic lesion edema growth (ΔNWU) after successful thrombectomy treatment. ΔNWU (OR 0.74, 95% CI 0.68 to 0.8, P<0.001) and CCC+ (OR 13.39, 95% CI 4.88 to 36.76, P<0.001) were independently associated with functional independence. Conclusion A comprehensive assessment of cerebral collaterals using the CCC model is strongly associated with edema growth and functional independence in acute stroke patients successfully treated by endovascular thrombectomy.

Background: Parenchymal hematoma (PH) is a major complication after endovascular treatment (EVT) for ischemic stroke. The hypoperfusion intensity ratio (HIR) represents a perfusion parameter reflecting arterial collateralization and cerebral microperfusion in ischemic brain tissue. We hypothesized that HIR correlates with the risk of PH after EVT. Methods: Retrospective multicenter cohort study of patients with large vessel occlusion who underwent EVT between 2013 and 2021 at one of the 2 comprehensive stroke centers (University Medical Center Hamburg-Eppendorf, Germany and Stanford University School of Medicine, CA). HIR was automatically calculated on computed tomography perfusion studies as the ratio of brain volume with time-to-max (Tmax) delay >10 s over volume with Tmax >6 s. Reperfusion hemorrhages were assessed according to the Heidelberg Bleeding Classification. Primary outcome was PH occurrence (PH+) or absence (PH−) on follow-up imaging. Secondary outcome was good clinical outcome defined as a 90-day modified Rankin Scale score of 0 to 2. Results: A total of 624 patients met the inclusion criteria. We observed PH in 91 (14.6%) patients after EVT. PH+ patients had higher HIR on admission compared with PH− patients (median, 0.6 versus 0.4; P <0.001). In multivariable regression, higher admission blood glucose (adjusted odds ratio [aOR], 1.08 [95% CI, 1.04–1.13]; P <0.001), extensive baseline infarct defined as Alberta Stroke Program Early CT Score ≤5 (aOR, 2.48 [1.37–4.42]; P =0.002), and higher HIR (aOR, 1.22 [1.09–1.38]; P <0.001) were independent determinants of PH after EVT. Both higher HIR (aOR, 0.83 [0.75–0.92]; P <0.001) and PH on follow-up imaging (aOR, 0.39 [0.18–0.80]; P =0.013) were independently associated with lower odds of achieving good clinical outcome. Conclusions: Poorer (higher) HIR on admission perfusion imaging was strongly associated with PH occurrence after EVT. HIR as a surrogate for cerebral microperfusion might reflect tissue vulnerability for reperfusion hemorrhages. This automated and quickly available perfusion parameter might help to assess the need for intensive medical care after EVT.

Background and purpose: Mechanical thrombectomy (MT) has proven to be the standard of care for patients with acute ischemic stroke due to large vessel occlusion (AIS‐LVO). However, high revascularization rates do not necessarily result in favorable functional outcomes. We aimed to investigate imaging biomarkers associated with futile recanalization, defined as unfavorable functional outcome despite successful recanalization in AIS‐LVO patients.Methods: A retrospective multicenter cohort study was made of AIS‐LVO patients treated by MT. Successful recanalization was defined as modified Thrombolysis in Cerebral Infarction score of 2b–3. A modified Rankin Scale score of 3–6 at 90 days was defined as unfavorable functional outcome. Cortical Vein Opacification Score (COVES) was used to assess venous outflow (VO), and the Tan scale was utilized to determine pial arterial collaterals on admission computed tomography angiography (CTA). Unfavorable VO was defined as COVES ≤ 2. Multivariable regression analysis was performed to investigate vascular imaging factors associated with futile recanalization.Results: Among 539 patients in whom successful recanalization was achieved, unfavorable functional outcome was observed in 59% of patients. Fifty‐eight percent of patients had unfavorable VO, and 31% exhibited poor pial arterial collaterals. In multivariable regression, unfavorable VO was a strong predictor (adjusted odds ratio = 4.79, 95% confidence interval = 2.48–9.23) of unfavorable functional outcome despite successful recanalization.Conclusions: We observe that unfavorable VO on admission CTA is a strong predictor of unfavorable functional outcomes despite successful vessel recanalization in AIS‐LVO patients. Assessment of VO profiles could help as a pretreatment imaging biomarker to determine patients at risk for futile recanalization.

To investigate whether unfavorable cerebral venous outflow (VO) predicts reperfusion hemorrhage after endovascular treatment (EVT), we conducted a retrospective multicenter cohort study of patients with acute ischemic stroke and large vessel occlusion (AIS-LVO). 629 AIS-LVO patients met inclusion criteria. VO profiles were assessed on admission CT angiography using the Cortical Vein Opacification Score (COVES). Unfavorable VO was defined as COVES ≤ 2. Reperfusion hemorrhages on follow-up imaging were subdivided into no hemorrhage (noRH), hemorrhagic infarction (HI) and parenchymal hematoma (PH). Patients with PH and HI less frequently achieved good clinical outcomes defined as 90-day modified Rankin Scale scores of ≤ 2 (PH: 13.6% vs. HI: 24.6% vs. noRH: 44.1%; p < 0.001). The occurrence of HI and PH on follow-up imaging was more likely in patients with unfavorable compared to patients with favorable VO (HI: 25.1% vs. 17.4%, p = 0.023; PH: 18.3% vs. 8.5%; p = <0.001). In multivariable regression analyses, unfavorable VO increased the likelihood of PH (aOR: 1.84; 95% CI: 1.03–3.37, p = 0.044) and HI (aOR: 2.05; 95% CI: 1.25–3.43, p = 0.005), independent of age, sex, admission National Institutes Health Stroke Scale scores and arterial collateral status. We conclude that unfavorable VO was associated with the occurrence of HI and PH, both related to worse clinical outcomes.