Human Visual Transduction Mechanism-Inspired Adaptive Enhancement Network during Endoscopy

Area of Science:

• Medical imaging
• Artificial intelligence
• Endoscopy

Background:

• Disposable endoscopes reduce cross-contamination but have lower resolution due to smaller sensors.
• Lower resolution in disposable endoscopes degrades diagnostic performance for both subjective and automated analysis.
• Improved adenoma detection rates are linked to clearer endoscopic imaging.

Purpose of the Study:

• To propose an adaptive enhancement network (AEN) to enhance lesion details in disposable endoscope images.
• To improve the diagnostic performance of disposable endoscopes by recovering image resolution and clarity.
• To leverage artificial intelligence inspired by human visual mechanisms for enhanced medical imaging.

Main Methods:

• Developed an adaptive enhancement network (AEN) mimicking the human visual pathway.
• Incorporated rod cell modules (RCM) for shape-sensitive segmentation and cone cell modules (CCM) for detail-oriented enhancement.
• Utilized a framework analogous to photoreceptor layers, outer plexiform layer, inner nuclear layer, inner plexiform layer, and ganglion cell layer.

Main Results:

• The AEN effectively enhanced lesion regions, producing sharper edges and finer textures with reduced blur and artifacts.
• Experimental results demonstrated significant improvements in image quality for lesion detection.
• Comprehensive evaluations across four datasets confirmed the generalizability and effectiveness of the AEN.

Conclusions:

• The AEN successfully recovers details in low-resolution disposable endoscope images, aiding endoscopists.
• The proposed AI network offers real-time processing advantages for endoscopic diagnostics.
• This technology has the potential to improve diagnostic accuracy and patient outcomes in procedures using disposable endoscopes.