Image_Classification

Chromosome Analysis or Karyotyping is done to find

1. Abnormalities such as birth defects/genetic diseases etc.
2. Structural Inconsistencies

Laboratory process:

1. Take Sample of Your Cells
2. Sample placed in a laboratory dish that allows the cells to grow.
3. Stain the cells.
4. Observe Under a Microscope (Metaphase)

• Chromosomal suspension is dropped onto slides and stained which results in a kind of banding pattern (what we get is called a metaphase spread)

You should get 46 chromosomes. Two of these chromosomes are sex chromosomes (determine the sex of the person being tested). Females - XX, Males - XY The other 44 are called autosomes.

Problem Statement:

Grouping/Rearranging the chromsomes from the metaphase into 23+1 Classes (As Below)

Data Science Initiative:

1. Process Metaphase

2. Extract Individual Chromsomes (OPENCV - Python)

3. Orientation & Feature Extraction (Image Processing & Cleaning)

Features Extracted

Length	Area	Banding Profile	Centromere's Location (& Length of both arms )
in Pixels	in Pixels

4. Modeling (Machine Learning)

          Nested Support Vector Machines
                 - 17 Years of Historical Data of Patients ( Metaphases & Karyotyped Training Data)
                 - After Extracting features from the all chromosomes, feeding them into a two-layer support vector machines.
                        - First layer classifies them into 'bigger classes'.
   

   

                        - Second layer classifes the chromosomes from those 'bigger classes' to individual classes 1-24.
   

   

5. Prediction

          Achieved a Prediction Accuracy of 86% 
          Various other methods such as CNNs, Deep networks were tested.
          (Please contact me personally for code)