I need you to develop some software for me. I would like this software to be developed for Windows using Python.
I have a dataset composed of 2 classes with MRI samples of different patients. In the first class there are brain MRIs of patient with Alzheimer and in the other there are brain MRIs of cognitively normal patients. Every MRI sample is a sequence of pictures, like a video. That means that a MRI sample is a tridimensional matrix.
The dataset is already preprocessed, in particular I've extracted the brain from every initial MRI sample. I've taken all the samples from the OASIS-3 Database ([login to view URL]) and then i preprocessed them.
I've created two versions of the same dataset, using different pre-processing techniques and I've reached 70% accuracy on both datasets, but my convLSTM is really really trivial. One dataset has MRI samples with dimension 176*256*256 and the other has MRI samples with dimensions 91*109*91. You can test your neural network on both dataset, here the links:
[login to view URL]
[login to view URL]
What I need from you is a convLSTM classifier. In the paper i've attached some researchers created a convLSTM that classifies smoking/non-smoking patients from MRI samples with an accuracy of 96% (IT'S A PROBLEM REALLY SIMILAR TO MINE). I would like you to use the same structure of the convLSTM they used in this paper, because it proved to work very well with MRI samples. In particular, I cite from the paper: "In this study, a 3-layer LSTM model was used, and the central 100 slices were applied as the input data. A 1,024-dimensional feature set was extracted from each slice using GoogleNet. The last layer of the ConvLSTM model was a full-connected layer."
And i need an explanation of the structure of your convLSTM as well as an explanation of the parameters you have selected. In particular, why, in your opinion, the convLSTM you've created performs well with MRI samples.
The goal for this project is a classification accuracy on the test dataset at least of 80%
Is all clear ? Thanks so much