Lung Cancer

Leading cause of millions of deaths

Survivability is high if detected earlier

Difficult to diagnose early

Use of CT images for diagnosis

Non-obstructive image extraction

Current methods detect lung cancer nodules at later stages

There is a need to detect it early

Contributions

First phase

Second Phase

Selection of features using Linear Discriminative Analysis

Classification using Modified Gravitational Search Algorithm

Enhancement and Extraction Methods

Filtering and Enhancement

Feature Extraction

Texture

Wavelet-based

Histograms

Used for noise removal

Prevalent in medical images

Use of filters if neighboring pixels are near 0 or 255

Use of histogram equalization to improve contrast

Plotting through histograms and determine gray levels

Variance

Mean

Standard Deviation

Skewness

Kurtosis

For shape distribution and determines anomalies in images

Determines number of gray level fluctuations

Average gray level in regions

Based on tails

Negative

Positive

Spatial relationship of gray level pixels

Outputs 22 features

Homogeneity

Energy

Entropy

Contrast

Correlations

Maximum values as maximum shape

Quantity of data

For compression

Determines values of minor constrasts

Variance of pixels based on neighbors

Linear dependence of gray levels

Transformation to frequencies through linear transformation

Feature Reduction

LDA

Combines bad features into one

Classification Methods

Deep Belief Networks

Restricted Boltzmann Machine

Modified Gravitational Search Algorithm

For parameter learning

Helps in learning parameters

Stochastic method

For weight optimization

Weight initialization

Force evaluation

Fitness evaluation

Selection of random value

Mass and force updates

Use of maximum specificity

Updates values based on fit, best, and worst values

Note: lacks further explanations on why certain methods were used

Optimal solution and termination

Fine-tuning

Use of backpropagation

Results and Discussion

High accuracy and Specificity

LDA is a better than PCA and ICA

Note: Comparison of performance should be in tabular form than bar charts

S.K., L., Mohanty, S. N., K., S., N., A., and Ramirez, G. Optimal
deep learning model for classification of lung cancer on ct images. Future
Generation Computer Systems 92 (2019), 374–382.