Evaluation Algorithm for the Effectiveness of Stroke Rehabilitation Treatment Using Cross-Modal Deep Learning

Area of Science:

• Medical imaging analysis
• Machine learning in healthcare
• Neurorehabilitation

Background:

• Accurate evaluation of stroke rehabilitation treatment effects is crucial for optimizing patient care.
• Existing methods face challenges with positron emission tomography (PET) image restoration and evaluation data recognition.

Purpose of the Study:

• To develop an advanced evaluation algorithm for stroke rehabilitation treatment effects.
• To improve the accuracy and efficiency of assessing patient recovery using multimodal imaging data.

Main Methods:

• A cross-modal deep learning approach was employed, utilizing magnetic resonance imaging (MRI) and PET scans from stroke patients.
• A three-dimensional cyclic adversarial neural network was used to restore missing PET data based on MRI-PET mapping.
• Multifeature fusion integrated RGB, depth, gray, and normal images from both MRI and PET for comprehensive analysis.

Main Results:

• The proposed algorithm demonstrated accurate restoration of PET images.
• High recognition accuracy for evaluation data, exceeding 95%, was achieved.
• The algorithm provided a high evaluation accuracy for stroke rehabilitation treatment effects with rapid processing times (0.56–0.91s).

Conclusions:

• The developed algorithm offers a robust and accurate method for evaluating stroke rehabilitation.
• The cross-modal deep learning approach shows significant potential for practical application in clinical settings.
• This technology can aid in optimizing stroke treatment plans based on precise recovery assessments.