Bioinformatics – Speaker Notes :-

1. what are u saying in the slide

2. flow of the slide
3. what are we concluding in the slide- what would it lead to.

Slide 12 :
In this slide we explain the brief outline and approach in Alzheimer disease
prediction from MRI image .

The process is to read the image , preprocess and normalize the image ,
segmentation of brain image into 3 components , combine newly extracted features
to the subject profile data for further treatment , here we apply PCA and do
classification using Naïve bayes classifier.

Now let us explain – how we done the preprocessing.

Slide 13 :

Here we explain the steps involved in preprocessing the image using FSL tool.

Preprocessing is done by splitting the brain image data into multiple equal size
boxes . Where each voxel represents spatial location of the brain.

Objective here is to do feature extraction by reducing the number of voxels. This is

achieved by segmenting brain image into 3 parts ( Grey matter , White matter and
CSF) initially.

In this process we have not only reduced the number of voxels but also segmented
the brain image into 3 parts – that can be used for further processing,

After segmentation – next step would be feature extraction and dimension reduction

Slide 14:
Since the original image comprises of 176X208X176 voxels . It is important to
identify those voxels that would not only help us is disease identification.

To achieve this we normalize the data and then segment the image into 3 parts to
create statistical parametric map in identifying those voxels that are different in
intensity and excluding voxels with same intensity.

This is an integral step in dimension reduction as we have now reduced the number
of voxels to 121X145X121 without loosing any critical information in disease
prediction.
Slide 15:

Upon segmentation and dimension reduction of the image data , it is important to

important to combine this data with the subject profile data for data consolidation .

This is achieved by segmenting brain image into Grey matter , White matter and
CSF and adding these components to the subject profile data.

This consolidated data is a critical step in feature extraction and classification.

Slide 16 :

During the data preparation process – we have some nominal , ordinal and
categorical data.
Hence while data processing we need to keep in mind – which fields require
normalization and which donot.

Slide 17 :

Upon normalizing the data – we select a random sample and split the sample into
training data (70%) and test data (30%).

Since Age and Gender are strong indicator of Alzheimer disease – we classify the
data into different levels of severity based on CDR and MMSE scores combined with
Age group for effective classification and early detection.

Naïve Bayes Classification algorithm was most effective as compared to other

classification techniques .

Slide 18 :

To validate the data – we randomly select sample of size 85 from a population of

216 and split each of the sample to train (70%) and test (30 %).

We have repeated this iterative process of selecting random sample of size 85 from
a population of 216 , eight (8) times. This process is called cross validation.

The results of the iterative process helps us create the confusion matrix to test the
accuracy of the prediction with the actual data.

We have achieved an accuracy of 97 %.

Slide 19 :
This slide reflects the actual Vs predicted data classification based on our sub
segmentation of the data mentioned in step 17.

A confusion matrix would help us understand the performance of our classification

model.

Step 20 :
We have total 65 subjects in training data containing 36 non-demented and 29
demented subjects.

Naïve Bayes classification is giving the best accuracy compared to other

classification method.

The Confusion matrix for the testing data gives the following interpretation:
Accuracy = 96.9 %