GNS Healthcare
April 1, 2023
Angel-Legs M20®
April 1, 2023Use of Deep Learning to Predict Final Ischemic Stroke Lesions from Initial Magnetic Resonance Imaging JAMA Network Open, Neurology
Importance
Objectives
To determine whether a deep learning model can predict final infarct lesions using magnetic resonance images (MRIs) acquired at initial presentation (baseline) and to compare the model with current clinical prediction methods.
Design, Setting and Participants
In this multicenter prognostic study, a specific type of neural network for image segmentation (U-net) was trained, validated, and tested using patients from the Imaging Collaterals in Acute Stroke (iCAS) study from April 14, 2014, to April 15, 2018, and the Diffusion Weighted Imaging Evaluation for Understanding Stroke Evolution Study–2 (DEFUSE-2) study from July 14, 2008, to September 17, 2011 (reported in October 2012). Patients underwent baseline perfusion-weighted and diffusion-weighted imaging and MRI 3 to 7 days after baseline. Patients were grouped into unknown, minimal, partial, and major reperfusion status based on 24-hour imaging results. Baseline images acquired at presentation were inputs, and the final true infarct lesion at 3 to 7 days was considered the ground truth for the model. The model calculated the probability of infarction for every voxel, which can be thresholded to produce a prediction. Data were analyzed from July 1, 2018, to March 7, 2019.
Main Outcomes and Measures...
Area under the curve, Dice score coefficient (DSC) (a metric from 0-1 indicating the extent of overlap between the prediction and the ground truth; a DSC of ≥0.5 represents significant overlap), and volume error. Current clinical methods were compared with model performance in subgroups of patients with minimal or major reperfusion.
Results
The deep learning model prediction is more stable across all subgroups with a mean volume difference closer to zero than ADC and Tmax predictions. The line inside the box represents the median volume difference. The boundaries of boxes represent the 25th and 75th percentile of volume difference. The error bar represents upper and lower 95% confidence intervals.
Conclusions
The deep learning model has successfully predicted infarct lesions from baseline imaging without reperfusion information and achieved comparable performance to existing clinical methods. Predicting the subacute infarct lesion may help clinicians prepare for decompression treatment and aid in patient selection for neuroprotective clinical trials.
Relevance to Healthcare Field
According to the Centers for Disease Control and Prevention, stroke is the fifth leading cause of death and a major cause of serious long-term disabilities in the United States. Up to 92% of strokes in the US are ischemic; worldwide, 15 million people suffer from stroke each year. Although controlling the risk factors to prevent an acute ischemic stroke is the goal, reperfusion therapy is the only effective treatment to reverse ischemic changes. This study used a specific type of deep convolutional neural network architecture known as a U-net to predict final infarct lesions in ischemic stroke patients using baseline Magnetic Resonance Images (MRIs) as inputs. It was able to predict three to seven-day infarctions. The results from the study can be compared with the current processes used involving the diffusion-perfusion mismatch paradigm. The proposed model outperformed existing methods in patients with minimal and major reperfusion for Positive Predictive Value (PPV)/specificity and sensitivity, respectively. It is crucial to be able to predict each patient’s outcome once the first image using the U-net is obtained, so management can prepare for decompression surgery and osmotherapy appropriately, if needed, and select patients for future clinical trials for neuroprotective agents.