Predicting Stroke Damage with MRI and Neural Networks
Author Information
Author(s): Bagher-Ebadian Hassan, Jafari-Khouzani Kourosh, Mitsias Panayiotis D., Lu Mei, Soltanian-Zadeh Hamid, Chopp Michael, Ewing James R.
Primary Institution: Henry Ford Hospital
Hypothesis
We hypothesized that, given a T2WI at the chronic stage of stroke (3 months post-stroke), an ANN might be trained to directly predict the size and pattern of the tissue recovery from the information available in acute phase MR images.
Conclusion
The trained ANN can accurately predict the size and pattern of ischemic lesions in stroke patients using acute-phase MRI data.
Supporting Evidence
- The ANN produced maps of predicted outcomes that were well correlated with the T2WI at 3 months.
- The study used a leave-one-out cross-validation method for training and testing the ANN.
- The ANN achieved an Area Under the Receiver Operator Characteristic Curve (AUROC) of 0.89.
Takeaway
Doctors can use special computer programs to look at brain scans and guess how much damage a stroke has done to the brain, helping them make better treatment decisions.
Methodology
An Artificial Neural Network (ANN) was trained using acute-phase MRI images to predict the chronic T2WI outcome at 3 months post-stroke.
Potential Biases
The ANN may confuse closely located lesions and CSF areas, leading to misinterpretation.
Limitations
The study had a small sample size and potential errors due to image mis-registration and quality.
Participant Demographics
Mean age was 56.25 years, with various ischemic stroke subtypes including cardioembolism and large vessel atherosclerosis.
Statistical Information
P-Value
p<0.0001
Statistical Significance
p<0.0001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website