Chronic Ulcers Healing Prediction through Machine Learning Approaches: Preliminary Results on Diabetic Foot Ulcers...

Area of Science:

• Medical Informatics
• Computational Biology
• Wound Healing Research

Background:

• Diabetic foot ulcers affect millions globally, necessitating improved prediction of healing trajectories.
• Accurate prognosis is vital for effective treatment and complication reduction in chronic wound care.

Purpose of the Study:

• To enhance diabetic foot ulcer prognosis and clinical decision-making through predictive modeling.
• To compare the efficacy of various Deep Neural Network (DNN) and Machine Learning (ML) models for wound healing prediction.

Main Methods:

• Utilized a dataset of 1766 diabetic foot wounds with at least three follow-up visits.
• Incorporated clinical features: Wounds Bed Preparation (WBP) scores, wound area, depth, and tissue status.
• Evaluated 12 predictive models, including a three-layer Long Short-Term Memory (LSTM) recurrent DNN.

Main Results:

• The LSTM DNN model achieved 80% accuracy, with an AUC of 0.85, recall of 0.80, precision of 0.79, and F1-score of 0.80.
• Key predictors identified: initial wound depth and area, followed by second-visit wound area and granulation tissue percentage.
• Model performance was highest when trained on wound data from four visits.

Conclusions:

• Predictive modeling offers a valuable tool for managing diabetic foot ulcers, supporting clinical practice.
• Future work includes developing a semantic segmentation model for tissue classification (necrosis, slough, granulation) to further enhance prediction accuracy.