Deep learning-based detection and segmentation of diffusion abnormalities in acute ischemic stroke

Abstract

Background: Accessible tools to efficiently detect and segment diffusion abnormalities in acute strokes are highly anticipated by the clinical and research communities. Methods: We developed a tool with deep learning networks trained and tested on a large dataset of 2,348 clinical diffusion weighted MRIs of patients with acute and sub-acute ischemic strokes, and further tested for generalization on 280 MRIs of an external dataset (STIR). Results: Our proposed model outperforms generic networks and DeepMedic, particularly in small lesions, with lower false positive rate, balanced precision and sensitivity, and robustness to data perturbs (e.g., artefacts, low resolution, technical heterogeneity). The agreement with human delineation rivals the inter-evaluator agreement; the automated lesion quantification of volume and contrast has virtually total agreement with human quantification. Conclusion: Our tool is fast, public, accessible to non-experts, with minimal computational requirements, to detect and segment lesions via a single command line. Therefore, it fulfills the conditions to perform large scale, reliable and reproducible clinical and translational research.

Left Hemisphere Bias of NIH Stroke Scale Is Most Severe for Middle Cerebral Artery Strokes

“…The lesion core was defined in DWI, in combination with the Apparent Diffusion Coefficient maps (ADC) by two experienced evaluators and was revised by a neuroradiologist until reaching a final decision by consensus. Further details are in our previous publication ( 16 ).…”

Automatic comprehensive aspects reports in clinical acute stroke MRIs

“…The feature analysis enriches the AI models, increasing their interpretability and their potential usefulness. Therefore, our system ADS 13 is suited to output not only the predicted ASPECTS but also the feature vector (AFV) showing the proportion of each brain region affected by the infarct (as in Figs. 3 and 4), the graphic representation of how the pre-trained model interprets the AFV components to predict scores in each region (Supplementary Fig.…”

“…The ASPECTS atlas defines the 10 areas considered in the ASPECTS system: the caudate, the lentiform, the internal capsule (IC), the insula, and the cortical / subcortical regions from M1-M6 24 . This proposed ASPECTS deformable 3D atlas is publicly available in ADS 13 . The visual ASPECTS rating was done by two evaluators, and finally defined by consensus with a neuroradiologist.…”

“…The most efficient models were Random Forest (RF), Multi-layer Perceptron (MLP), and Support Vector Machine (SVM). These three best models were assembled to generate a final model that is included in our deployed pipeline to calculate ASPECTS in ADS 13 . The accuracy to predict the total ASPECTS was inferior of that to predict regional scores, for both humans and machine.…”

“…Supplementary Fig. 3 illustrates how the pre-trained RF model in ADS 13 interprets the contribution of each AFV component to predict ASPECTS in a given region.…”