A novel SpaSA based hyper-parameter optimized FCEDN with adaptive CNN classification for skin cancer detection

Area of Science:

• Dermatology
• Medical Imaging
• Artificial Intelligence

Background:

• Skin cancer is a prevalent global health concern, with early detection significantly impacting treatment outcomes.
• Image preprocessing is crucial for enhancing the quality of dermoscopic images in automated skin cancer diagnosis.
• Existing methods require further optimization for accuracy and efficiency.

Purpose of the Study:

• To compile accessible preprocessing techniques for skin cancer imaging.
• To propose a novel method for segmenting and classifying skin lesions using deep learning and optimization algorithms.
• To provide a starting point for researchers in automated skin cancer diagnosis.

Main Methods:

• A fully convolutional encoder-decoder network combined with the Sparrow search algorithm (FCEDN-SpaSA) was developed for image segmentation.
• A neighbor-based search strategy was integrated into SpaSA, balancing exploration and exploitation.
• An adaptive convolutional neural network (CNN) technique was employed for classifying skin lesions.

Main Results:

• The FCEDN-SpaSA method achieved high segmentation accuracies: 95.28% (ISBI 2017), 95.89% (ISIC 2018), and 92.70% (PH2 dataset).
• Classification accuracy reached 91.67%.
• The proposed approach demonstrated comparable classification performance to incremental learning techniques with reduced energy, storage, memory access, and training time.

Conclusions:

• The developed FCEDN-SpaSA method offers an efficient and accurate approach for skin lesion segmentation and classification.
• This technique shows potential for improving automated skin cancer diagnosis systems.
• The study highlights the effectiveness of integrating advanced AI algorithms for medical image analysis.