International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 09 | Sep 2024
p-ISSN: 2395-0072
www.irjet.net
Towards Robust Skin Cancer Diagnosis: Deep Fusion of VGG16 and Mobile Net Features Anirudh Kuchibhotla1 SRM University1 ---------------------------------------------------------------------***--------------------------------------------------------------------and deep learning (dl) help distinguish benign and malignant Abstract – Skin cancer has become a dangerous condition in skin lesions. While technology might be beneficial to healthcare professionals, it shouldn’t replace their training and expertise.
the modern world. It can be classified as either non-melanoma or melanoma (benign or malignant). Scars, dark spots, or changes in the skin appearance can all be indicators of skin cancer. Changes in irritability, size, form, or color could be indicators of skin cancer. This project concentrates on ensembling the extracted features from the vgg16 and MobileNet pre-trained models and fed to the ML models and a deep neural network. To enhance some current procedures and create modern approaches that would enable precise models that decrease the time gap between the diagnosis and treatment period in identifying skin cancer. With ensemble features from vgg16 and mobileNet pre-trained models, experimental findings show the effectiveness of the deep neural network and achieve an accuracy of 87%. Index
This study is focused on the detailed examination of skin lesions, including benign and malignant forms. The project’s goal is to create reliable diagnostic techniques by closely analyzing the various traits that these lesions display. It also tackles the problem of improving the way that skin cancer is currently diagnosed by fusing cutting-edge technology with machine learning methods.In particular, it concentrates on combining features that are taken from trained models like VGG16 and MobileNet. These features are then fed into a deep neural network and a variety of machine learning algorithms by utilizing ensemble techniques. Notably, the obtained accuracy of 87% indicates promising developments in the field of skin cancer detection and provides assurance for more effective and efficient diagnostic methods in the future.
Key Words:- Skin cancer, Malignant, Benign, Deep Neural Network,pre-trained models
1.INTRODUCTION
2.RELATED WORK
Millions of people worldwide lose their lives to skin cancer every year, which is one of the most common types of cancer. Each year, this condition affects about three million people in the United States alone.[6] This is more than the total number of cases of breast, lung, and colon cancer combined together. The main cause of skin cancer is UV radiation exposure. Physicians typically perform a biopsy in order to make a diagnosis. Skin cancer, which includes both benign and malignant types, is one of the most important health issues facing nowadays. Planning an appropriate diagnosis and course of treatment requires the ability to distinguish between tumors that are benign and malignant. Early warning signs of skin cancer may be visible changes in the look of the skin, such as dark patches, scars, or texture changes. There are two main types of skin cancer: benign and malignant. Melanoma is one example of a malignant skin cancer that develops anywhere and is derived from melanocytes.
The findings of the data regarding skin cancer is entailed by exploring 10 research papers from 2013 to 2023. Aman Kamboj et al.[7] proposed a methodology for melanoma skin cancer detection that incorporated artifact removal, segmentation, feature extraction, and classification algorithms. It employed preprocessing techniques such as artifact removal through thresholding and segmentation using K-means in the HSV color space. Feature extraction involved asymmetry and color features, with classification performed using SVM and KNN algorithms. This study underscored the importance of feature combination and machine learning techniques in achieving accurate melanoma detection, with a focus on enhancing diagnostic capabilities through a fusion strategy for segmentation. Evgin Goceri[3] discussed an Automated skin cancer detection that relies on dl methods for classifying and detecting lesions , with pre-processing techniques like DullRazor for hair removal. Architectures such as GoogleNet, VGG16, ResNet50, and DenseNet are merged to enhance accuracy in distinguishing benign and malignant lesions. Challenges included high computational costs, driving the need for faster training methods for mo bile device compatibility.Databases like ISIC, HAM10000, PH2, and ISBI2017 are utilized for training these models, emphasizing the importance of robust datasets in advancing skin cancer
They have the ability to spread, aggressive growth patterns, and invasion of nearby tissues. Benign skin tumors, on the other hand, usually grow more slowly and do not behave in an aggressive manner. A correct diagnosis is essential to medicine since it guides treatment choices and determines prognosis. This highlights the importance of distinguishing between various sorts. These days, machine learning (ml)
© 2024, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 884