Slide 76 -
Melanoma and skin cancers vs Image Processing 2 Skin cancer and melanoma Skin cancer : most common of all cancers
According to the latest statistics available from the National Cancer Institute, skin cancer is the most common of all cancers in the United states.
More than 1 million cases of skin cancer are diagnosed in the US each year.
What’s shown here are some examples of skin lesion images.
The four images shown on the left are various form of skin lesions, cancerous or non-cancerous.
The two on the right are a specific form of skin cancer: melanoma.
3 What is Melanoma? A type of skin cancer that starts from melanocytes
6th leading cause of cancer death in the US
No single etiology
Some risk factors include:
Sun exposure -depleting ozone layer
Presence of many or unusual moles
benign malignant skin 6 Skin cancer and melanoma Skin cancer : most common of all cancers
[ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ] Use of color to distinguish malignant and benign tumors Skin tumors can be either malignant or benign
Clinically difficult to differentiate the early stage of malignant melanoma and benign tumors due to the similarity in appearance
Proper identification and classification of malignant melanoma is considered as the top priority because of cost function
Classification of skin tumors using computer imaging and pattern recognition
Previous texture feature algorithms successfully differentiate the deadly melanoma and benign tumor seborrhea kurtosis
Relative color feature algorithm is explored in this research for differentiate melanoma and benign tumors, dysplastic nevi and nevus
Successfully classify 86% of malignant melanoma using relative color features, compared to the clinical accuracy by dermatologists in detection of melanoma of approximately 75%
Types of Melanoma Superficial Spreading Melanoma
70%, neck, legs, pelvis
15%, dome-shaped nodule
8 %, Common in dark-skin
Lentigo Maligna Melanoma
5 %, sun-exposed area, mistaken for age spot
1.7%, ½ amelanotic Benign vs Malignant 9 10 Automated Melanoma Recognition UsingImaging Techniques Melanoma is one of the most aggressive cancers, but it can be healed by surgical excision successfully only if it is recognized in the early stage.
Since the melanoma emerges as a tiny dot in the topmost skin layer, it can be examined during routine medical check up.
Although the lesions are accessible, in many cases, it is a difficult task to make decisions whether nevi are benign or malignant.
Further, frequent use of biopsy is also not encouraged.
Hence, to assist dermatologist's diagnosis, it is useful to develop an automated imaging-based melanoma recognition system. 11 Uncontrolled growth of melanocytes give rise to dark and elevated appearance of melanoma.
Neoplasm- growth of tissue, tumor
Melanoma is a type of malignant skin cancer that starts from melanocytes. It’s caused by uncontrolled growth of melanocytes that gives rise to tumor.
Nonetheless there are risks factors that highly attributed to its incidence. Some of the them are:
amount of sun exposure – the more cumulative exposure the higher
presence of many of unusual mole – people with many moles in the body
Fitzpatrick’s Skin Type I and II have higher risk –1975 Thomas Fitzpatrick, Harvard skin typing system based on skin complexion and response to sun exposure
genetic predisposition – if there history of melanoma that runs in the family
According to a study ,compared to general population, people who with 2 risk factors have 3.5 times risk of developing MM and 20 times those who have 3 or more risk factors.
These are the types of melanona
As you see, SSM is the most prevalent one that makes up 70% of most diagnosed melanoma
In this work, images of superficial spreading melanoma were only explored.
The reason being, and the problem that this work is trying to solve, Dysplastic Nevi ( a benign mole) has properties that are highly similar to this SSM melanoma, which makes the diagnosis of melanoma difficult.
Melanoma Incidence Incidence highest in Caucasian skin Graph one- Caucasian has the highest incidence of MM.
Having fair complexion is one of the risk factors. Researches attribute this to low level of melanin that absorbs harmful UV radiation in fair skin, thus UV penetrates much deeper layer affects the surrounding cells.
NCHS – national center for health statistics
Bureau of Health Statistics
Graph two – men shows higher incidence than women.
A study of in Germany linked this trend to mutation of genes called BRAF 4% and CDKN2A 1%.
Melanoma Incidence Incidence increases with age Graph thee – Incidence increases with age. Link to cumulative sun exposure
Some studies suggested that people who had significant exposure to UV at younger age have higher risk in later age when UV exposure decreases.
Age-adjusted- distribution of age by percentage
It’s a way of data normalization so that you can compare two different countries, cities and so forth
Need standard population distribution
Who use it
NCHS – national center for health statistics
Bureau of Health Statistics
What to say
So these are three graphs that show melanoma incidence in different dimensions: based on race, gender, and age.
Here, it’s evident that Melanoma has its favorites, so to speak.
Melanoma Incidence It is estimated that 62,480 men and women (34,950 men and 27,530 women) will be diagnosed with and 8,420 men and women will die of melanoma of the skin in 2008 (SEER) Surveillance Epidemiology and End Results
What to say
This is the combination of all of the data from the previous slides.
Average of 4.2 percent increase per year
Survival Rate by Stage The American Joint Committee on Cancer (AJCC) TNM System http://www.cancer.org The imaging is performed by a special CCD camera combined with an epiluminescence microscope in order to produce digitalized ELM images of the skin lesions.
Once the images are captured, the lesion has to be segmented from the background and useful information should be extracted from the lesion region.
Based on the extracted features, decisions have to be made about the nature of the skin lesion.
The decisions should be supported by descriptive justifications so that dermatologist can understand the decision making process. 22
Contact person: Assoc. Prof. PonnuthuraiNagaratnamSuganthan, email: email@example.comTel: 6790-5404
Collaborators: Prof. C L Goh, MD, National Skin Centre, Singapore & Dr. H Kittler, University of Vienna
This is an on-going project. We have implemented the segmentation, feature extraction and clasifcation modules satisfactorily, although further improvements are desirable. The module to provide explanations supporting the classifcation decisons is yet to be developed. siii 23 24 Skin cancer and melanoma Skin cancer : most common of all cancers
Melanoma : leading cause of mortality (75%) [ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ] Although represent only 4 percent of all skin cancers in the US, melanoma is the leading cause of mortality.
They account for more than 75 percent of all skin cancer deaths. 25 Skin cancer and melanoma Skin cancer : most common of all cancers
Melanoma : leading cause of mortality (75%)
Early detection significantly reduces mortality [ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ] The time line shown here is the 10 year survival rate of melanoma.
If caught in its early stage, as seen here, melanoma can often be cured with a simple excision, so the patient have a high chance to recover. Hence, early detection of malignant melanoma significantly reduces mortality.
26 [ Image courtesy of “An Atlas of Surface Microscopy of Pigmented Skin Lesions: Dermoscopy” ] Clinical View Dermoscopy view Dermoscopy Dermoscopy is a noninvasive imaging technique, and it is just the right technique for this task.
It has been shown effective for early detection of melanoma.
The procedure involves using an incident light magnification system, i.e. a dermatoscope, to examine skin lesions.
Often oil is applied at the skin-microscope interface.
This allows the incident light to penetrate the top layer of the skin tissue and reveal the pigmented structures beyond what would be visible by naked eyes.
27 28 Dermoscopy Dermoscopy improves diagnostic accuracy by 30% in the hands of trained physicians
May require as much as 5 year experience to have the necessary training
Motivation for Computer-aided diagnosis (CAD) of pigmented skin lesion from these dermoscopy images.
Clinical view Dermoscopy view
In the future, with the development of new algorithms and techniques, these computer procedures may aid the dermatologists to bring medical break through in early detection of melanoma.
29 40,000 people between 1988-2001
Cancer stage is categorized into TNM level
T – tumor ( localized)
N – regional lymph-nodes
The key point is the earlier the better of survival
5- and 10- year survival mean percentage of people who live at least 5 and 10 years respectively after being diagnosed
Diagnosis- ABCDE System E= evolution/elevation
What to say
ABCDE system is the tool for detecting melanoma. This is a list of criteria that can be used for distinguishing between benign and malignant melanocytic skin lesions.
A- if you draw a line across the center of MM, you’ll see that is not symmetric compared to regular mole
B- the border is uneven or ragged is a sign of melanoma
C-if there are multiple shades of pigment is presence
D- diameter > 6mm
Dermatologist adds E for either evolution if lesion changes upon observation or E for elevation.
Suspicious lesion is followed by histological confirmation. Where the problems lie Atypical nevi acquire several properties similar to melanoma, their recognition posed high difficulties even to experts. The classical ABCD guidance is not reliable therefore cannot be used as sole indicator for detection of melanoma for both clinical and public examination.
In clinical setting, recognition and discrimination are highly subjective with rate of success based on experts’ years of experience. As was found, inexperienced dermatologists showed decrease sensitivity in the detection of melanoma in both live and photo examinations.
General practitioner – 62% sensitivity and 63% specificity
Dermatologist – 80% sensitivity and 60% specificity OK, so we have the ABCD diagnosis tool plus the experts.
So anyone with sort of skin lesion can step in a clinic get the ABCD tool and experts examination undertaken then there you have the results.
You either have benign mole or malignant melanoma at the end of the consultation. Everything just goes as plan.
Unfortunately it is not always the case.
Sensitivity – TP/TP+FN
Specificity – TN/TN+FP
Read the bullet
The objective of the this work is to address these problems
MM and DN ABCD Rules Malignant Melanoma Dysplastic Nevi Here you have some samples of MM on the top row and DN on the bottom row
Atypical Nevi (mole) – shares some sometimes all characteristics of MM.
This actually what makes melanoma detection difficult.
Objectives To construct an automated, image-based system for classification of Malignant Melanoma and Dysplastic Nevi using solely the visual texture information of the lesion. The system will be based on methodologies that emanate and/or correlated with human vision therefore will closely emulates human experts only with greater extent of accuracy, reliability and reproducibility
Investigate new segmentation methods that will be effective on both lesions
Extract most relevant texture information from the image
Construct a classification system of the lesion
Ultimate goal is the construction of the classification system
The uniqueness of the system is the fact that :
only texture information is used – robust in color variability
Methodologies used through out the whole process emanate from the human vision thus emulate human expert
Systems, Materials and Tools Image database
Original tumor images
512x512 24-bit color images digitized from 35mm color photographic slides and photographs
160 melanoma, 42 dysplastic, and 80 nevus skin tumor images
Binary images drawn manually and reviewed by the dermatologist for accuracy
Computer vision and image processing tools developed at our research lab
Statistical analysis tools CVIPtools Other approach - Texture System for Melanoma Detection Outline of the Process Here you have the outline of the process
Each of the subsequent step is dependent of the of the preceding steps. In other terms, the results of subsequent step is only as good as the results of preceding steps.
Therefore, since segmentation is the top most of the hierarchy, its important to make sure the method is robust.
Hypotheses Due to observable pattern disruption in the skin tissue driven by the MM, It is hypothesize that measuring magnitude of pattern disruption provides discriminative features for diagnosing MM.
Since visual texture is highly length-scale dependent, It is hypothesized that the detection and analysis methods that explore texture at different scales such as the wavelet is the most appropriate approach.
It is hypothesized that texture descriptors that emanate from and highly correlated with human vision system provide the utmost representation, and thus yield a more contextual system—a system that closely emulate human expert Item one – skin has distinct uniform pattern (glyphic pattern).
MM disrupts texture.
Quantifying texture differences between MM and NV is more reliable method than color-based ( color-based in prone to variability in imaging system)
Item two – texture come in different sizes.
Detection method that explore texture image at different possible scale is more sensitive than methods that are using one scale.
Example of this snake-based ( gradient-based), Normalized Cut, histogram threshold
Item three – there are many texture descriptors that are purely algorithmic that may not necessary correlate with human vision.
One example is first-order statistics of texture ( variance ,mean), structure-based approach, laplacain of Gaussian.
Texture classifiers that emanate from or highly correlated with human visual system provides a closer approximation of experts perception of texture.
Visual Texture Texture Texture is regarded as what constitutes a macroscopic region. Its structure is simply attributed to the repetitive patterns in which elements or primitives are arranged according to a placement rule(Tamura et al, 1978).
Texture is both the number and types of its (tonal) primitive and their spatial arrangement (Haralick ,1979).
The term texture generally refers to repetition of basic texture elements called texels. The texel contains several pixels, whose placement could be periodic, quasi-periodic, or random. Natural textures are generally random, whereas artificial textures are often deterministic or periodic. Texture may be course, fine, smooth, granulated, rippled, regular, irregular, or linear (Jain, 1989).
Texture is intuitively viewed as descriptor in providing a measure of properties such as smoothness, coarseness, and regularity (Gonzales and Woods, 1990).
Texture is an attribute representing the spatial arrangement of the gray levels of the pixels in a region (IEEE, 1990).
Texture is both grey level of a single pixel and its surrounding pixels, which was coined as a unit texture, texels. These texels conformed repetitive patterns that dictated the effective texture analysis approach (Karu et al, 1996).
Patterns which characterize objects are called texture in image processing (Jähne, 2005). Technical Definition Texture has no single definition.
Definitions from previous literature dedicated in studying texture
The first three definitions, tells us texture is composed of a building block that is spatially arranged based on the placement rule (periodic, quasi periodic, or random): like a brick a single brick is the building block, the arrangement of the bricks that gives rise to a texture of a brick wall
Texture is descriptors for smoothness, coarseness, and regularity
In computer vision
Spatial arrangement of gray levels of the pixel
Texture and Human Vision System Pre-attentive visual system-1962-1981
Disproved conjecture that second-order is processed in the vision system
Terminator-end of lines, corners
Elongated blobs of different sizes - granularity As one of the hypothesis. Texture characterization emanate from visual system closely emulates experts
Neuroscientist, studied perception of texture
Before disproving, he conjectured that second-order statistics is processed in the vision system, and He claimed that two textures with similar second-order statistic is not pre-attentively recognizable.
In other words without close inspections, two different texture with same sec stat would seem to look similar.
After series of experiments, he finally suggested that textons are the major player for texture discrimination.
And the textons are contrast, terminators. granularity
Texture discrimination Second-order statistics Textons Textons instead of second-order statistics that cause the texture discrimination The image on left is an example of two different textures with the same SO that is not pre-attentively detectable.The right image is two different textures with the same SO but pre-attentively detectable. Among others this leads to the final statement texture discrimination is made possible through the textons. Here in this one is the difference termination of the two texture elements .In this work, the second-order statistics CoM and contrast of edge elements will be explored for extracting visual texture properties of skin lesion. Texture and Human Vision System Frequency and Orientation
Multi-frequency and orientation analysis
decomposition (1968) –Campbell and Robson
Simple cells of the visual cortex respond to narrow ranges of frequency and orientation, cells act as 2D spatial filter-(1982) De valois et al.
Orientation-based texture segregation involves the generation of a neural representation of the surface boundary whose strength is nearly independent of the magnitude of orientation contrast - Motoyoshi and Nishida (2001)
More studies had been conducted in part to understand human vision.
This Campbell and Robson found that when signal received by the eye is decomposed into multiple frequencies and orientation
Another work in the subsequent year that further support the previous finding that simple cells are highly selective/tuned to narrow frequency and orientation.
Another work found that neural representation of texture boundary is formed that is independent of magnitude and orientation of the contrast
In this work in wavelet analysis will be used for segmentation. Frequency and contrast
Texture Method Design Creation of relative color images
Segmentation and morphological filtering
Relative color feature extraction
Design of tumor feature space and object feature space
Establishing statistical models from relative color features COLOR Create Relative Color Skin Tumor Images Purpose
to equalize any variations caused by lighting, photography/printing or digitization process
to equalize variations in normal skin color between individuals
the human visual system works on a relative color system
Mask out non-skin part in the image to calculate the normal skin color
Separate tumor from the image
Remove the skin color from the tumor to get a relative color skin tumor image
CVIPtools functions were used to create relative color skin tumor images
Calculate Skin Color Tumor Image Relative Color Tumor Image Segmentation and Morphological Filtering Image segmentation was used to find regions that represent objects or meaningful parts of objects
Morphological filtering was used to reduce the number of objects in the segmented image
Easy to use CVIPtools for experimenting and analysis
Feature Extraction Relative Color Feature Extraction Necessary to simplify the raw image data into higher level, meaningful information
Feature vectors are a standard technique for classifying objects, where each object is defined by a set of attributes in a feature space.
Totally 17 color features and binary features were extracted using CVIPtools
The three largest objects, based on the binary feature ‘area’, were used in feature extraction
Histogram features, that is, color features, were extracted in each color band from relative color image objects 17 Features Binary features
Thinness Histogram features in R, G, B bands
17 Features (Cont.) Design Two Feature Spaces Tumor feature space
consists of 277 feature vectors correspond to 277 skin tumor images.
each feature vector has 51 feature elements, which are the total of 17 features of each three largest objects within the same tumor.
Object feature space
had 842 feature vectors corresponding to 842 image objects
each feature vector has 17 feature elements, which were the binary features and color features stated as above Establishing Statistical Models Two feature spaces serve as two data models in order to maximize the possibility of success
Two classification models, Discriminant Analysis and Multi-layer Perceptron, were developed for both data models
The training and test paradigm is used in statistical analysis to report unbiased results of a particular algorithm
due to small size of data set, 282 images, we used the leave x out method, with both one and ten for x
Partek software was used
to analyze the data representing the features
to develop a model or rules for classifying the tumors
07/13/2005 Computer Vision and Image Processing Research Lab @ ECE Dept., SIUE Quadratic Discriminant Analysis A statistical pattern recognition technique based on Bayesian theory, which classifies data based on the distribution of measurement data into predefined classes
Normalization the feature data as preprocessing
performed to maximize the potential of the features to separate classes and satisfy the requirement of the modeling tool such as Quadratic discriminant analysis for a Bayesian distribution of the input data
Variable selection was used to choose dominant features. Multi-Layer Perceptron A feed forward neural network
neural networks modeled after the nervous system in biological systems, based on the processing element the neuron
widely used for pattern classification, since they learn how to transform a given data into a desired output.
Principal Component Analysis (PCA) as preprocessing
a popular multivariate technique, is to reduce dimensionality by extracting the smallest number components that account for most of the variation in the original multivariate data and to summarize the data with little loss of information
the dispersion matrix selected for PCA in this project is correlation
Multi-Layer Perceptron (Cont.) Creation, training and testing of neural networks
Creation a neural network involves selection of hidden and output neuron types and a random number generation.
Four output neuron types – Softmax, Gaussian, Linear and sigmoid
Three hidden neuron types – Sigmoid, Gaussian and Linear
Scaled Conjugate Gradient algorithm is used for learning in this project.
Automated and independent of user parameters
Avoids time consuming
Stopping criteria, sum-squared error, is selected to determine after how many iterations the training should be stopped
The trained data is then tested on itself first to examine how far the neural network is able to classify the objects correctly.
Leave x partition out method is used for testing the algorithm