Department of Computer Science and Engineering Faculty of Science and Information Technology Daffodil International University. We hereby declare that this project was carried out by us under the supervision of Al Amin Biswas, Lecturer (Senior Scale), Department of CSE, Daffodil International University. We are deeply grateful and indebted to Al Amin Biswas, Lecturer (Senior Scale), Department of CSE Daffodil International University, Dhaka.
The first step is image processing with subcontinental data and the second step is the use of machine learning and deep learning based on the algorithm "Convolutional Neural Network" which can easily help to detect those samples that have skin disease or not in a very low - cost way. This system can be used by any class of people who cannot afford the high quality and expensive test method that compares those algorithms, as a result we probably achieved 83.86% accuracy using CNN for detecting different types of skin disease with disease name. Skin disease is a big problem in the world today and one-fourth of the world's people are affected by this disease.
Skin diseases can be harmful to detect at an early stage due to lack of medical resources, proper treatment, proper equipment and lack of medical acumen. If it is possible to diagnose skin diseases at an early stage, this disease can be cured very easily. Therefore, to overcome this obstacle, we are going to come up with a method that can easily predict skin diseases through image processing.
CNN has the best accuracy between CNN and SVM, so we did our skin disease prediction with CNN.
Motivation
Objective
Daffodil International University 3 through image processing and they can easily understand what disease they are infected with.
Report Layout
BACKGROUND
Literature Review
Daffodil International University 5 best accuracy using CNN which accuracy rate is too high than 99.00% but their method is too time consuming. They basically act on the ultraviolet rays from the sun that cause skin cancer. Ultraviolet effects on our skin directly because the sun kisses us directly on the outside. They focused on a naive bias algorithm where they got an accuracy rate of 90.00% where the data set was small or large, which is not a fact.
They have applied a very low cost method that can be used by any classes of people worldwide. They inspired us to develop a new M-health solution that differentiates between normal and abnormal images. This inspired us to develop a new M-Health solution that differentiates between normal skin images and abnormal images using mobile neural networks.
Therefore, in the classification of skin diseases using image processing and SVM by Kumar et al. They also focused on level data and proposed image-based data to help support the vector machine learning algorithm. They applied this by programming python They proposed RGB format and transform into 1-dimensional array to set their database.
6] used a methodology to create a phototype to detect skin disease using the basis of CNN. Other author describes in skin disease prediction that their method detects any type of skin disease using CNN. They focused on user interface system where the affected area's picture is given and then the result shown with the disease name.
Normal and affected skin sometimes look the same, so use the sharpening filter function to improve the contrast of the edge of the image. In this case, they also helped themselves with CNN, but they did not show the exact result. They tested various diseases without precision and computer vision extracted the order from the image combined with CNN.
Comparison
In the first proposal, he proposed a global skin disease diagnostic system using a deep convolutional neural network (CNN) [20], which achieved a Top-1 accuracy of 73.1% and a Top-5 accuracy of 91. 0% are reached. In 2016, together with other researchers, he proposed a second work [21], in which they proved that the use of characteristics of skin lesions facilitates the diagnosis of skin diseases, because many diseases are so similar in visual aspects. Kawahara et al. in [16] used the MobileNet network trained on a public library DermoFit and classified the skin lesions into ten categories.
Limitation of existing system
METHODOLOGY
- Proposed system of Skin Disease Classification
- Convolutional Neural Network
- Support Vector Machine
- K-Nearest Neighbor
- Naive Bayes
- Dataset Information
Where it processes the actual data and then it will generate some modules with precision. It then performs the three-layer work in this system which has a convolutional layer, a pooling layer and a fully connected layer. First, few pixels of the input images enter the first convolution + ReLu (rectified linear unit) where the convolutional part filters the images and ReLu converts nonlinear to 0 and speeds up the training time.
Then it goes through the next layer, which is the merging layer, where the size of the images is dimensionally reduced. The final layer has 5 neurons, which are nail fungus, skin allergy, hair follicles, acne and also normal skin. Daffodil International University 12, which adopted a decision plan that divided between the given facilities and classified them.
SVM receives the main size of the input image and then resizes these pixel images for implementation. Using some feature extraction techniques like dropout, smooth layer, pooling layer, dense layer, maxpool2D etc. to find the accurate classification of the image. But to detect which types of skin diseases are affected, K-NN helps in simple classification.
The purpose of the five types of skin disease data, as the disease name suggests, is to classify whether a given image contains acne, allergies, nail fungus, hair loss, or normal skin. When we know something about image classification first, we understand perspective variation, scale variation, warping, intra-class variation, etc. It simply identifies the unknowns by finding the most common class among K closet example to visualize this process five types of data within each respective category are relatively related together in the dimensional space.
Like when we set the normal and abnormal image, it labels the image and says the given images are normal or abnormal. The class or label of the image is predicted as a result of creating the probability, distributions of all the data that is shown, then it decides which one is associated with its exact disease name When it found a similarity, it follows two things for prediction. Considering all the given conditions as an image of skin allergy, given a series of characteristics, it can predict and observe skin allergy.
After collecting the data set first, we preprocess all the data according to the unique disease name such as Melanocytic Nevi (NV), Basal Cell Carcinoma (BCC), Melanoma, Molluscum Warts and other viral infections, benign lesions keratosis-like (BKL), Psoriasis photos Lichen Planus and related diseases, Seborrheic Keratoses and other Benign Tumors, Ringworm Candidiasis Tinea and other fungal infections, eczema and atopic dermatitis. Daffodil International University 15 conducted our experiment around images on a dataset with infected areas and also normal area 77% of training data and 23% of testing data were applied to our method.
RESULT ANALYSIS
Experimental Result
Model Description
When we enter a skin disease image into our system, the system can register which skin disease the image corresponds to and can name the disease.
Impact on Society, Environment and Sustainability
- Impact on Society
- Impact on Environment
- Ethical Aspects
- Sustainability Plan
The use of machine learning for skin disease detection raises several ethical concerns, such as data privacy, bias in training data, and liability for misdiagnosis. Data privacy is a major concern in any machine learning application involving personal health information. In the case of skin disease detection, the images used to train and test the algorithms may contain sensitive information that can be used to identify individuals.
It is important to ensure that all data is properly anonymised and stored securely to prevent unauthorized access. Another ethical issue with machine learning-based skin disease detection is the potential for bias in the training data. Algorithms are only as good as the data they're trained on, so if the training data is skewed in any way, the algorithms are likely to make predictable predictions as well.
This could lead to unequal treatment of certain populations, such as marginalized or minority groups. Machine learning algorithms are not perfect and there is always the possibility of a wrong diagnosis, which can have serious consequences for the patient. It is important to ensure that clear guidelines are in place to deal with misdiagnosis and that those responsible for developing and deploying these systems are held accountable for their actions.
This includes reducing its energy consumption and carbon footprint, addressing ethical concerns about privacy and bias, and ensuring its long-term financial sustainability while benefiting patients, healthcare providers and society. The plan must consider the interconnected impacts of the technology to ensure its long-term sustainability.
FUTURE WORKS AND CONCLUSION
Summary of the Study
Conclusion
Implication for Further Study
Abid, “Malignant melanoma classification using deep learning: datasets, performance metrics, challenges and opportunities,” IEEE Access , vol. 21] Haofu Liao, Yuncheng Li, and Jiebo Luo, “Skin thickness classification versus skin lesion characterization: Achieving robust diagnosis using multi-label deep neural networks,” in 2016 23rd International Conference on Pattern Recognition (ICPR), pp.
