A Robust Deep Learning Method for Predicting Brain Strokes Using the Innovative Densenet Algorithm in Comparison to Long Short-Term Memory (LSTM)

The brain strokes are serious medical conditions that demand quick diagnosis and intervention if one wishes to minimize the disabilities that come with them in order to give patients a better chance of recovery. With technological advancement in imaging techniques, particularly MRI, there is an opportunity to harness data from MRI images into a more correct stroke prediction. Traditional prediction methods using MRI images lack accuracy and efficiency and many times require more sophisticated algorithms. This research tries to increase the efficiency and accuracy of the prediction of a brain stroke by using MRI images and adopting the revolutionary method of Densely Connected CNN, known as DenseNet, comparing the performance with the LSTM algorithm. The two different groups involved in this study in order to assess the prediction algorithms for brain stroke diagnosis: Group 1-LSTM method; Group 2-DenseNet algorithm. The sample size was a total of 20, divided into two groups of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 0}$</tex> samples each. The algorithms were developed and then run in Python. Using some rigorous statistical testing, the sample size, using an alpha of 0.05 and a beta of 0.2, the resultant statistical power was 0.8. This was estimated on the basis of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 5 \%}$</tex> confidence from the findings. The excellent accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 2. 5 0 \%}$</tex> came from the DenseNet approach, whereas in the LSTM algorithm, this value was only <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 7. 5 \%}$</tex>. By statistical test, the <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{p}$</tex>-value is <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 0 0 1}(p&lt;0.05)$</tex>; thus, there is a significant difference between the accuracies obtained by the two algorithms; hence, DenseNet is more accurate.