KEY TAKEAWAYS
- The study aimed to identify genetic and metabolic factors contributing to deadly cancers like melanoma to improve global screening and prevention strategies.
- Customized CNN demonstrated superior accuracy in melanoma detection compared to alternative techniques based on comparative analysis.
Advancements in understanding genetics and metabolism have unveiled various abnormalities linked to cancer, a potentially fatal condition that can affect any part of the body. Among the most common forms is skin cancer, and its incidence is steadily increasing worldwide. Skin cancer includes subtypes such as squamous cell carcinoma, basal cell carcinoma, and the highly aggressive melanoma, responsible for the majority of fatalities. Given its prevalence and severity, regular screening for skin cancer is imperative.
Devika Moturi and the team conducted the study that aimed to emphasize the importance of skin cancer screening due to its increasing prevalence and potentially deadly nature, driven by a better understanding of genetic and metabolic factors underlying cancer.
The study involved the utilization of deep learning techniques, specifically MobileNetv2 and DenseNet, for the rapid and precise detection of skin cancer. It focused on analyzing the HAM10000 dataset, which comprised 10,000 images of skin lesions, encompassing both nonmelanocytic and melanocytic tumors.
The primary objective was to distinguish between malignant and benign tumors using these techniques. Comparative evaluations were conducted to assess the performance of each method, leading to the inference of results based on their respective performances.
The results revealed that following model evaluation, MobileNetV2 achieved an accuracy of 85%, while the customized CNN exhibited a higher accuracy of 95%. Subsequently, a web application has been constructed using the Python framework, offering a user-friendly graphical interface integrated with the best-performing model.
Through this interface, users can input patient information and upload lesion images. The uploaded images were then classified utilizing the corresponding trained model, enabling the prediction of whether the image depicts cancerous or non-cancerous conditions. Furthermore, the web application provides the percentage of cancer involvement for the identified lesions.
According to the comparative analysis, the customized CNN demonstrated superior accuracy in detecting melanoma compared to MobileNetV2. No funding was allocated for this study.
Source: https://pubmed.ncbi.nlm.nih.gov/38407684/
Moturi, D., Surapaneni, R.K. & Avanigadda, V.S.G. Developing an efficient method for melanoma detection using CNN techniques. J Egypt Natl Canc Inst 36, 6 (2024). https://doi.org/10.1186/s43046-024-00210-w.
