SKIN DISEASE DETECTION AND MULTI-CLASS CLASSIFICATION USING CONVOLUTION NEURAL NETWORK MODEL

Abstract

Skin diseases are a global health issue that require early and accurate diagnosis for treatment. A Convolutional Neural Network (CNN) model is used to develop an advanced skin disease detection and multi-class classification system. The proposed system automates dermatological image-based skin disease classification using deep learning. A diverse dataset of images of skin conditions like dermatitis, psoriasis, eczema, and melanoma is collected and curated. This dataset trains and validates the CNN model, which automatically learns relevant features and patterns from input images. The CNN architecture is optimized for skin disease classification. Convolutional layers extract hierarchical features, and pooling layers downsample spatially, simplifying computation. Fully connected layers also influence classification. Data augmentation is used during training to improve model generalization. This reduces overfitting and lets the model handle skin conditions and image characteristics. The proposed CNN model is evaluated using accuracy, precision, recall, and F1 score. The developed system accurately identifies and classifies skin diseases, according to experiments. The proposed model outperforms traditional methods, suggesting it could be used in healthcare. This automated skin disease detection system could help dermatologists diagnose patients faster and more accurately, improving patient outcomes and reducing healthcare costs.

Specification

Description:The python code is a Django web application for skin disease prediction and user registration/authentication
Imports: The code imports various libraries and modules necessary for the application, including Django modules for web development, machine learning libraries like Keras and scikit-learn for model training and evaluation, and others like OpenCV and NumPy for image processing and data manipulation.
Global Variables and Constants: It defines some global variables like cnn_algorithm and class_labels which are used throughout the application.
Views: Views are the functions that handle web requests and return web responses. In this code, there are several views like DiseasePrediction, DiseasePredictionAction, runCNN, Signup, UserLogin, etc. Each view corresponds to a different page or action in the web application.
Disease Prediction: The DiseasePrediction and DiseasePredictionAction views seem to be responsible for handling the process of predicting skin diseases based on input images. The DiseasePrediction view renders a form where users can upload images, and the DiseasePredictionAction view processes the uploaded image, performs disease prediction using a pre-trained convolutional neural network (CNN) model, and displays the result.
Convolutional Neural Network (CNN) Model: The code includes functionality for training and using a CNN model for skin disease classification. The runCNN view trains the CNN model if it's not already trained, or loads the pre-trained model if it exists. It then evaluates the model and displays performance metrics such as accuracy, precision, recall, and confusion matrix.
User Registration and Authentication: The views Register, Signup, Login, and UserLogin handle user registration and authentication. Users can register with a username, password, contact details, email, and address. Upon registration, the user data is stored in a MySQL database. The UserLogin view verifies user credentials during login.
Rendering Templates: The views render HTML templates using the render function, passing context data to the templates for dynamic content generation. In this invention we are using CNN to classify skin diseases from images as CNN gain lots of success and popularity in the field of image classification. To train CNN we have used skin disease dataset which contains 9 different types of diseases such as 'Actinic Keratosis', 'Basal Cell Carcinoma', 'Dermatofibroma', 'Melanoma', 'Nevus', 'Pigmented Benign Keratosis', 'Seborrheic Keratosis', 'Squamous Cell Carcinoma' and 'Vascular Lesion'. After training CNN algorithm, we can upload any test image then CNN will detect and classify disease from that image. Fig. 4.1 shows the block diagram of proposed system.
This module is used for user registration. In this module user must upload the username, password, contact number, email ID, address. After entering the details then press the register button. In this module user has to enter their username and password to access the account.
According to the facts, training and testing of any deep neural network or transfer learning involves in allowing every source image via a succession of convolution layers by a kernel or filter, rectified linear unit (ReLU), max pooling, fully connected layer and utilize SoftMax layer with classification layer to categorize the objects with probabilistic values ranging from [0,1].
Convolution layer as is the primary layer to extract the features from a source image and maintains the relationship between pixels by learning the features of image by employing tiny blocks of source data. It's a mathematical function which considers two inputs like source image I(x,y,d) where x and y denotes the spatial coordinates i.e., number of rows and columns. d is denoted as dimension of an image (here d=3, since the source image is RGB) and a filter or kernel with similar size of input image and can be denoted as F(k_x,k_y,d).
The output obtained from convolution process of input image and filter has a size of C((x-k_x+1),( y-k_y+1),1), which is referred as feature map. Let us assume an input image with a size of 5×5 and the filter having the size of 3×3. The feature map of input image is obtained by multiplying the input image values with the filter values.
, C , C , Claims:
1. We claim to achieve superior accuracy in detecting and classifying various skin diseases compared to traditional diagnostic methods, potentially reaching or exceeding benchmark performance metrics.
2. We claim By utilizing a comprehensive and diverse dataset, the framework claims to generalize well across different skin types, tones, and demographics, ensuring fair and accurate predictions for all patients.
3. We claim to provide real-time predictions, enabling healthcare professionals to receive immediate diagnostic feedback and facilitating prompt clinical decision-making.
4. We claim to offer an intuitive and accessible interface that allows healthcare providers to easily upload images and interpret results without requiring extensive technical knowledge.
5. We claim by automating the detection process, the model claims to significantly reduce the time needed for skin disease diagnosis, allowing dermatologists to focus on patient care.

Documents