Automatic Diagnosis of Skin Cancer

According to WHO one out of three cancer diagnoses worldwide is skin cancer. In Karachi, Pakistan according to the statistics provided by Dow diagnostic research and references laboratory (DDRRL) out of all register cancer cases from 2010-2015, 5.1% cases were registered as non-melanoma skin ca

Project Title

Automatic Diagnosis of Skin Cancer

Project Area of Specialization

Artificial Intelligence

Project Summary

According to WHO one out of three cancer diagnoses worldwide is skin cancer. In Karachi, Pakistan according to the statistics provided by Dow diagnostic research and references laboratory (DDRRL) out of all register cancer cases from 2010-2015, 5.1% cases were registered as non-melanoma skin cancer and 0.3% cases were register as melanoma skin cancer for both genders. According to the Shaukat Khanum memorial cancer hospital and research center (SKMCH&RC) statistics in 2017, out of all cancer diagnosis case, skin cancer is the 8th deadliest cancer with the number of register cases 3.85% for both genders. 

Basically, there are different types of skin cancer, but they are usually divided into two categories melanoma and non-melanoma. In recent decades, melanoma’s incidence and the mortality rate have increased dramatically, thus becomes a major problem in public health. However, early-detected patients have a higher chance of curing, especially if the cancer is detected in its early stages and removed, the cure rate can be over 90%.

The manual inspection from dermoscopy images made by dermatologists is usually time-consuming, error-prone and subjective even well-trained dermatologists may produce widely varying diagnostic results.

To overcome this issue we are building ADSC web application. Automatic Diagnosis of Skin Cancer (ADSC) web application will facilitate dermatologist in diagnosing different types of skin cancer fast and error-free, bypassing dermoscopy image of skin lesions.

The result that dermatologist will get after passing images will help him/her to start treatment of patients early and to take any immediate step required for that particular type of skin cancer. As melanoma skin cancer type is the deadliest one so identifying melanoma cancer in a patient could help him to survive as dermatologists would be able to start treatment early.

Project Objectives

Automatic diagnosis of skin cancer (ADSC) is research and web-based application that will predict skin cancer. ADSC will use deep learning techniques to classify different types of skin cancer by taking dermoscopy skin lesion image as an input. It will help dermatologist to identify the different type of skin cancer by just passing dermoscopy images of the lesion.

the main objective of this project is to mitigate the trouble faces by a dermatologist for early and error-free detection of deadliest skin cancer like melanoma. so that the treatment of patient start early which could save many lives.

Project Implementation Method

We will be using a publicly available dataset on skin cancer mainly from ISIC-ARCHIVE. We will use the python programming language for training and testing our dataset. the project will be build using deep learning approach, as it falls in the category of computer vision especially image recognition so we will use the Convolutional neural network for training our model. we may use already trained models of CNN like VG16 etc or we may try to build our own model based on time and resources we have. we may use flask or django framework to integrate our model to cloud base web application.

the project requires a high-end machine with advance GPU of Nvidia to train our model or as an alternate cloud service could be utilized to train our model whose price varies for a number of images and time required to train our model. for images in thousand, it will take several hours or days to train our model.  

Benefits of the Project

The manual inspection from dermoscopy images made by dermatologists is usually time-consuming, error-prone and subjective, even well-trained dermatologists may produce widely varying diagnostic results. ADSC web application is expected to mitigate the trouble faces by a dermatologist for early and error-free detection of deadliest skin cancer like melanoma. ADSC research results will further provide a deep analysis of how different algorithm with different parameters can be benefited in future skin cancer classification. If our research finding will produce a considerable result, then it will be beneficial for the artificial intelligence community and biomedical industry.

Technical Details of Final Deliverable

It requires a high-end machine to train our model and build project but once the model is trained and deployed on cloud base web application then it could be accessed by any machine.

to use the product a user needs:

1-stable internet connection.

2-updated web browser.

3-dermoscopy images.

images other than dermoscopy will not produce accurate results so it is important for the user to pass only dermoscopy image in the web app for prediction.

the web application will work fine in all major web browsers as long as the user has updated web browser installed on his system.

although any user whether he/she is a dermatologist or not can use this system it is highly recommended that only dermatologist will use this system and come up with the conclusion that whether a patient is suffering from deadliest skin cancer like melanoma or not. 

Final Deliverable of the Project

HW/SW integrated system

Type of Industry

Health

Technologies

Artificial Intelligence(AI)

Sustainable Development Goals

Good Health and Well-Being for People

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com