Artificial intelligence-based PRO score assessment in actinic keratoses from LC-OCT imaging using Convolutional...

Area of Science:

• Dermatology and Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• The histological PRO score grades actinic keratoses (AK) malignant potential by assessing dermal-epidermal junction (DEJ) undulation.
• Line-field confocal optical coherence tomography (LC-OCT) enables non-invasive, real-time PRO score quantification.
• Automated PRO score quantification using AI, specifically Convolutional Neural Networks (CNNs), on LC-OCT data is a potential advancement.

Purpose of the Study:

• To train and validate an AI model for automated PRO score quantification of AK using LC-OCT images.
• To assess the accuracy of AI-computed PRO scores against expert visual consensus.
• To evaluate the feasibility of this AI tool for clinical application in AK management.

Main Methods:

• Convolutional Neural Networks (CNNs) were trained to segment LC-OCT images of healthy skin and AK.
• PRO score models were developed based on histopathological standards.
• The AI models were trained on 237 LC-OCT AK images and tested on 76 images, comparing AI results to expert consensus.

Main Results:

• Significant agreement (75%) was found between AI-automated grading and expert visual consensus.
• AI-automated grading showed the highest correlation with visual scores for PRO II (84.8%), followed by PRO III (69.2%) and PRO I (66.6%).
• Misinterpretation occurred in 25% of cases, primarily due to DEJ shadowing and disruptive features like hair follicles.

Conclusions:

• CNNs demonstrate utility for automated PRO score quantification in LC-OCT images of AK.
• This AI approach offers a potentially feasible tool for clinicians to assess PRO scores during AK follow-up.
• Further refinement may be needed to address limitations like shadowing and follicular artifacts for improved accuracy.