Automated Skin Lesion Classification for Cancer

Project Title

Automated Skin Lesion Classification for Cancer

Project Area of Specialization Artificial IntelligenceProject Summary

Skin cancer Is the most common type of skin cancer and easy to operate if classify in early stage.Most of the time it is ignored and just treated as normal lesion which make it most deadly cancer our aim is to make an automated lesion classifier for classification of skin cancer.we will use more than one clinical approved dataset to trained the machine learning models than by using performance matrix we will find the model having the best accuracy.To make it available for general public to check weather they have cancer or not we will make android app,so they can take a picture and classify skin lesion keeping it as simple as possible    

Project Objectives

1. To propose an automated system,for classification of skin cancer in an efficient way.
   1. High accuracy
   2. Poratablity
   3. Cost free application
2. To propose a pre-processing technique to improve the image quality against low contrast.
3. To propose a tecnnique to solve unbalance data
   1. RandAugment strategy (contrast, brightness and color manipulations, zoom in/out ,rotation, translations, and cutout )
   2. Redcuing the number of classes
4. To propose regularization tecnniques
   1. L1/L2 regularization
   2. Early Stopping
   3. Dropout
5. Dataset Augmentation

Project Implementation Method

1. Preeprocssing the dataset
2. Spliting the dataset into
   1. Training set
   2. Testing set
   3. Validation set
3. Training the model with some hyperparamenter
   1. epochs
   2. learning-rate
   3. cross-validation
   4. loss function 
4. Making predication and evaluate the model by
   1. Confusion matrix
   2. Accuracy
   3. Precision
   4. Recall
   5. Specificity
   6. F1 score
   7. Precision-Recall or PR curve
   8. ROC (Receiver Operating Characteristics) curve
   9. PR vs ROC curve.

1. Optimizing the model
   1. Hyperparameter tuning
   2. Adrdressing Overfiting
   3. Data Augmentation
   4. Droupout
2. Repeating the proccess untill it achived max accuracy
3. Making an Android Application to run the model and perform classification using Mobile AI (Not Server or Cloud)

Benefits of the Project

Skin cancer Is the most common type of skin cancer and easy to operate if classify in early stage.Most of the time it is ignored and just treated as normal lesion which make it most deadly cancer our aim is to make an automated lesion classifier for classification of skin cancer.we will use more than one clinical approved dataset to trained the machine learning models than by using performance matrix we will find the model having the best accuracy.To make it available for general public to check weather they have cancer or not we will make android app,so they can take a picture and classify skin lesion keeping it as simple as possible    

Melanoma is a serious form of skin cancer that begins in cells known as melanocytes. While it is less common than basal cell carcinoma (BCC) and squamous cell carcinoma (SCC), melanoma is more dangerous because of its ability to spread to other organs more rapidly if it is not treated at an early stage.

1. The estimated five-year survival rate for U.S. patients whose melanoma is detected early is about 99 percent.

1. An estimated 6,850 people (4,610 men and 2,240 women) will die of melanoma in the U.S. in 2020. 

Technical Details of Final Deliverable

An Android application which a have model inside of it for classification user can click a picture and can choose a picture form memory and can classify it from that application.The application will result a predictions about picture with some accuracy 

Final Deliverable of the Project Software SystemCore Industry ITOther Industries Medical Core Technology Artificial Intelligence(AI)Other Technologies OthersSustainable Development Goals Good Health and Well-Being for PeopleRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	70000
gtx 1080	Equipment	1	60000	60000
gtx 1080 power supply	Equipment	1	10000	10000