Convolutional Neural Network Models for Visual Classification of Pressure Ulcer Stages: Cross-Sectional Study

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Wound Care Technology

Background:

• Pressure injuries (PIs) present significant health challenges and economic burdens.
• Accurate staging of PIs is critical for effective treatment but remains difficult due to varied clinical presentations and lack of objective diagnostics.
• Deep learning, particularly Convolutional Neural Networks (CNNs), shows promise for improving classification accuracy in complex medical imaging, including wound assessment.

Purpose of the Study:

• To evaluate the efficacy of various CNN models (AlexNet, VGGNet16, ResNet18, DenseNet121) for the automated staging of pressure injuries.
• To develop an intelligent tool to assist healthcare professionals in the accurate and efficient staging of PIs.

Main Methods:

• Collected 853 raw pressure injury images across six stages (I, II, III, IV, unstageable, suspected deep tissue injury).
• Augmented the dataset 9 times through cropping and flipping, resulting in 7677 images.
• Trained AlexNet, VGGNet16, ResNet18, and DenseNet121 models on the augmented dataset, divided into training, validation, and test sets (8:1:1 ratio).

Main Results:

• DenseNet121 achieved the highest overall accuracy at 93.71% for PI staging.
• ResNet18, AlexNet, and VGGNet16 demonstrated overall accuracies of 92.42%, 87.74%, and 82.42%, respectively.
• The study confirmed the strong classification performance of CNN models on pressure injury images.

Conclusions:

• CNN-based models exhibit significant potential for developing highly efficient and intelligent PI staging systems.
• These AI tools could aid in standardizing PI assessment and potentially improve patient outcomes.
• Future research should compare these models against clinical expertise, such as nurses with varying experience levels, to validate real-world clinical utility.