OM-NAS: pigmented skin lesion image classification based on a neural architecture search

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computational Biology

Background:

• Manual design of convolutional neural networks (CNNs) for pigmented skin lesion classification demands significant expertise and extensive parameter tuning.
• Automating CNN architecture design is crucial for improving efficiency and accessibility in medical image analysis.

Purpose of the Study:

• To develop an automated approach, macro operation mutation-based neural architecture search (OM-NAS), for constructing CNNs tailored for pigmented skin lesion image classification.
• To enhance the accuracy and efficiency of classifying skin lesions using deep learning.

Main Methods:

• An improved, cell-oriented search space incorporating micro and macro operations (e.g., InceptionV1, Fire modules) was utilized.
• An evolutionary algorithm with macro operation mutation iteratively modified cell structures, analogous to viral DNA injection, to optimize CNN architecture.
• The optimal CNN architecture was assembled by stacking the best-discovered cells and evaluated on the HAM10000 and ISIC2017 datasets.

Main Results:

• The developed OM-NAS approach successfully constructed a CNN that demonstrated competitive or superior performance compared to state-of-the-art methods like AmoebaNet and InceptionV3+Attention.
• The automated CNN achieved average sensitivities of 72.4% on the HAM10000 dataset and 58.5% on the ISIC2017 dataset.
• The method significantly reduces the need for manual intervention in CNN design for skin lesion classification.

Conclusions:

• The macro operation mutation-based neural architecture search (OM-NAS) provides an effective automated solution for designing high-performance CNNs for pigmented skin lesion classification.
• This approach offers a promising direction for advancing automated medical image analysis, reducing reliance on expert knowledge.
• The validated performance on benchmark datasets underscores the potential of OM-NAS in clinical diagnostic support systems.