A comprehensive analysis of classification methods in gastrointestinal endoscopy imaging.

Area of Science:

• Medical imaging analysis
• Computer vision in healthcare
• Gastroenterology research

Background:

• Gastrointestinal (GI) cancers are a leading cause of death, with early detection crucial for survival.
• Current endoscopic surveillance has a high missed rate for early GI cancer precursors, partly due to human factors.
• Automatic GI disease classification systems are needed to flag suspicious findings and reduce diagnostic errors.

Purpose of the Study:

• To analyze the performance of computer vision methods in GI endoscopy across three major challenges (MediaEval 2017, 2018, 2019).
• To establish a benchmark for multi-class endoscopic image classification and encourage development of clinically applicable approaches.
• To identify challenges, shortcomings, and clinical credibility of current automated endoscopic analysis methods.

Main Methods:

• Comprehensive analysis of 21 participating teams' methods across Medico GI challenges (2017-2019).
• Evaluation of computer vision techniques applied to multi-class endoscopic images.
• Performance assessment based on metrics like Mathew correlation coefficient (MCC) and computational speed.

Main Results:

• Significant improvement in maximum MCC from 82.68% (2017) to 95.20% (2019).
• Demonstrated increase in computational speed of automated systems over the three years.
• Detailed insights into the strengths and weaknesses of various computer vision approaches for GI endoscopy.

Conclusions:

• Computer vision methods show substantial progress in classifying GI endoscopic findings, enhancing diagnostic accuracy.
• The analyzed challenges provide a valuable benchmark for developing reliable automated systems for clinical use.
• Further research is needed to address remaining shortcomings and ensure the clinical translation of these technologies.