An Improved Detection Algorithm for Ischemic Stroke NCCT Based on YOLOv5.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Cerebral stroke (CS) is a complex condition with diverse causes.
• Ischemic stroke (IS), a major subtype, involves disrupted cerebral blood flow.
• Noncontrast computed tomography (NCCT) is crucial for IS detection, but feature identification in image analysis is challenging.

Purpose of the Study:

• To develop an improved algorithm for detecting ischemic stroke (IS) in NCCT images.
• To enhance the selection of IS features within computational image analysis.
• To improve the accuracy and reliability of automated IS detection.

Main Methods:

• Proposed AC-YOLOv5, an adaptive local region contrast enhancement algorithm integrated with YOLOv5.
• The algorithm amplifies IS features in NCCT images before detection.
• Tested on two datasets with seven control experiments comparing against existing algorithms.

Main Results:

• AC-YOLOv5 achieved high accuracy rates of 94.1% and 91.7%.
• The algorithm demonstrated strong recall rates of 85.3% and 88.6%.
• Experimental results confirmed the algorithm's excellent accuracy, robustness, and generalizability.

Conclusions:

• AC-YOLOv5 significantly improves IS detection from NCCT images.
• The adaptive contrast enhancement effectively highlights IS features.
• The algorithm shows strong potential for clinical application in stroke diagnosis.