Adaptive Deep Video Frame Interpolation for Robust Laparoscopic Visualization: A Comparative Evaluation

Information from the abstract

ABSTRACT High‐quality temporal visualization is essential in laparoscopic surgery, where rapid instrument motion, nonrigid tissue deformation, illumination variation, and partial occlusions can reduce the quality of surgical video sequences and hinder accurate visual interpretation. Although deep learning‐based video frame interpolation (VFI) has achieved remarkable performance on natural image benchmarks, its applicability to laparoscopic surgical videos remains insufficiently investigated. This study presents a comprehensive evaluation of four state‐of‐the‐art deep learning‐based VFI models, namely AdaptiveConv, Revisiting Adaptive Convolutions (RAC), DAIN, and RIFE, for laparoscopic video enhancement. The models were systematically evaluated on two publicly available surgical datasets, GLENDA and ITEC SurgicalActions160, under experimental settings with and without data augmentation. Performance was assessed using objective image quality metrics, including Peak Signal‐to‐Noise Ratio (PSNR), Structural Similarity Index (SSIM), Root Mean Square Error (RMSE), and Learned Perceptual Image Patch Similarity (LPIPS), together with illustrative qualitative visual comparison, paired‐samples t tests, and evaluation across two independently evaluated datasets. Experimental results demonstrate that RAC consistently achieved the best overall performance, attaining a PSNR of 40.78 dB and an SSIM of 0.96 while simultaneously producing lower RMSE and LPIPS values than the other evaluated methods. Qualitative comparisons further showed that RAC more effectively preserved anatomical structures, tissue textures, and surgical instrument boundaries under challenging laparoscopic imaging conditions. Statistical analysis using paired tests with Holm–Bonferroni correction confirmed statistically significant differences between RAC and the competing methods across the evaluated metrics. Evaluation across the two independent datasets demonstrated consistent performance under diverse surgical procedures and imaging conditions. The findings establish a comprehensive benchmark for evaluating deep learning‐based VFI methods in laparoscopic surgery and provide evidence to support the selection and further evaluation of interpolation models for surgical video enhancement, computer‐assisted surgery, and surgical education.

Why this record is monitored

This record has an Impact Signal of 82/100 based on recency, source, collaboration, and bibliographic signals. It prioritizes monitoring and is not a judgment of research quality.

Related topics: Advanced Image Processing Techniques · Image Enhancement Techniques · Image and Video Quality Assessment