A New Multi-Layer Machine Learning (MLML) Architecture for Non-invasive Skin Cancer Diagnosis on Dermoscopic Images

Artificial intelligence (AI) has significantly impacted the healthcare industry, enabling the development of advanced medical devices and software that provide efficient and precise treatments. Health 4.0, the incorporation of computing and AI technologies into healthcare, is driving the industry's digital transformation and improving the diagnosis and treatment of diseases. AI can help detect diseases such as cancer at an early stage. AI can also lower the healthcare costs by reducing the need for unnecessary biopsies and speeding up the diagnostic process. Machine learning algorithms are commonly utilized in AI-powered healthcare studies and are also used in image-based research to diagnose a variety of diseases since the integration of AI into healthcare holds great potential to improve patient care and reduce costs. In this study, we present a multi-layer machine learning (MLML) method based on the joint use of machine learning algorithms to improve the success of skin cancer diagnosis. In this respect, the MLML method with 3 layers is proposed. In the first layer, decision tree, random forest, neural network, naive bayes, and support vector machine algorithms are used. After executing this layer, 5 different classification results are transferred to the second layer where k-nearest neighbor algorithm is utilized. In the last layer, the results are improved using the linear regression algorithm. Thanks to our method, images in the input dataset are classified into three groups: cancer, not cancer, and early-stage cancer. The multi-layer architecture is utilized to make joint decisions with different machine learning algorithms and remove the limitations of each algorithm so that more accurate decisions can be made. Fourteen feature extraction algorithms that were not previously used in skin cancer images are employed. Inclusion of age, gender, and region of the lesion in the decision-making process in addition to image features also contributes to obtaining better classification results. The performance of the proposed method was evaluated using four metrics. The conducted experiments showed that the MLML technique achieved 88.81% accuracy, 88.89% precision, 99.17% recall, and 93.75% F1-score in classifying skin cancer images. Finally, the results were compared with other relevant studies in the literature to demonstrate the superiority of the proposed method.

Dergi

JOURNAL OF ELECTRICAL ENGINEERING & TECHNOLOGY

Yıl

2024

Cilt

19

Sayı

4

Sayfa

2739-2755

ISSN

1975-0102

eISSN

2093-7423

DOI

10.1007/s42835-023-01758-8

WoS UID

WOS:001141873200001