2. Problem
Preventable medical error is a big killer.
In the US alone, 400,000 people die every year due to
avoidable medical error in hospitals - this is equivalent
to TWO JUMBO JETS crashing every single day!
--
NHS sets aside 26.1 Billion Dollars to settle outstanding
negligences and liabilities in clinical safety.
3. Causes of Avoidable Medical Errors
Procedures and training methods not reformed,
so mistakes happen again and again.
4. Features of the Dataset - Labelled 699 clinical cases
Nine real-valued features are chosen for each cell nucleus:
a) radius (mean of distances from center to points on the perimeter)
b) texture (standard deviation of gray-scale values)
c) perimeter
d) area
e) smoothness (local variation in radius lengths)
f) compactness (perimeter^2 / area - 1.0)
g) concavity (severity of concave portions of the contour)
h) concave points (number of concave portions of the contour)
i) symmetry
9. KNN - non parametric model to verify classification
Scaled feature vector
Identified precise k value using elbow method
For k = 3
We had an F1 score
0.95 for the anomaly
class
10. Unsupervised Anomaly Detection using SVM - Gaussian Kernel Trick
1)Objective is to train a one class svm gaussian
hypersphere that quarantines the benign cells.
2)Dropped labels from dataset and is split into benign and
malignant datasets.
3)Benign dataset is used to train the model.
4)Malignant dataset, the dataset that contains the outliers
is used to test.
5)A single class SVM is trained with a low gamma value,
that captures the influence of training examples on
classification.
11. Gaussian Distribution for benign and malignant cells
Benign multivariate gaussian Malignant multivariate
gaussian