Methodology to develop machine learning algorithms to improve performance in gastrointestinal endoscopy

Assisted diagnosis using artificial intelligence has been a holy grail in medical research for many years, and recent developments in computer hardware have enabled the narrower area of machine learning to equip clinicians with potentially useful tools for computer assisted diagnosis (CAD) systems. However, training and assessing a computer's ability to diagnose like a human are complex tasks, and successful outcomes depend on various factors. We have focused our work on gastrointestinal (GI) endoscopy because it is a cornerstone for diagnosis and treatment of diseases of the GI tract. About 2.8 million luminal GI (esophageal, stomach, colorectal) cancers are detected globally every year, and although substantial technical improvements in endoscopes have been made over the last 10-15 years, a major limitation of endoscopic examinations remains operator variation. This translates into a substantial inter-observer variation in the detection and assessment of mucosal lesions, causing among other things an average polyp miss-rate of 20% in the colon and thus the subsequent development of a number of post-colonoscopy colorectal cancers. CAD systems might eliminate this variation and lead to more accurate diagnoses. In this editorial, we point out some of the current challenges in the development of efficient computer-based digital assistants. We give examples of proposed tools using various techniques, identify current challenges, and give suggestions for the development and assessment of future CAD systems.

Favorable lifestyle before diagnosis associated with lower risk of screen-detected advanced colorectal neoplasia

AIM

To investigate the association between adherence to health recommendations and detection of advanced colorectal neoplasia (ACN) in colorectal cancer (CRC) screening.

METHODS

A total of 14832 women and men were invited to CRC screening, 6959 in the fecal immunochemical test arm and 7873 in the flexible sigmoidoscopy arm. These were also sent a self-reported lifestyle questionnaire to be completed prior to their first CRC screening. A lifestyle score was created to reflect current adherence to healthy behaviors in regard to smoking, body mass index, physical activity, alcohol consumption and food consumption, and ranged from zero (poorest) to six (best). Odds ratios (ORs) and 95%CIs were calculated using multivariable logistic regression to evaluate the association between the single lifestyle variables and the lifestyle score and the probability of detecting ACN.

RESULTS

In all 6315 women and men completed the lifestyle questionnaire, 3323 (53%) in the FIT arm and 2992 (47%) in the FS arm. This was 89% of those who participated in screening. ACN was diagnosed in 311 (5%) participants of which 25 (8%) were diagnosed with CRC. For individuals with a lifestyle score of two, three, four, and five-six, the ORs (95%CI) for the probability of ACN detection were 0.82 (0.45-1.16), 0.43 (0.28-0.73), 0.41 (0.23-0.64), and 0.41 (0.22-0.73), respectively compared to individuals with a lifestyle score of zero-one. Of the single lifestyle factors, adherence to non-smoking and moderate alcohol intake were associated with a decreased probability of ACN detection compared to being a smoker or having a high alcohol intake 0.53 (0.42-0.68) and 0.63 (0.43-0.93) respectively.

CONCLUSION

Adopted healthy behaviors were inversely associated with the probability of ACN detection. Lifestyle assessment might be useful for risk stratification in CRC screening.

Inter-observer agreement in the assessment of endoscopic findings in ulcerative colitis

Background

Endoscopic findings are essential in evaluating the disease activity in ulcerative colitis. The aim of this study was to evaluate how endoscopists assess individual endoscopic features of mucosal inflammation in ulcerative colitis, the inter-observer agreement, and the importance of the observers' experience.

Methods

Five video clips of ulcerative colitis were shown to a group of experienced and a group of inexperienced endoscopists. Both groups were asked to assess eight endoscopic features and the overall mucosal inflammation on a visual analogue scale. The following statistical analyses were used; Contingency tables analysis, kappa analysis, analysis of variance, Pearson linear correlation analysis, general linear models, and agreement analysis. All tests were carried out two-tailed, with a significance level of 5%.

Results

The inter-observer agreement ranged from very good to moderate in the experienced group and from very good to fair in the inexperienced group. There was a significantly better inter-observer agreement in the experienced group in the rating of 6 out of 9 features (p < 0.05). The experienced and inexperienced endoscopists scored the "ulcerations" significantly different. (p = 0.05). The inter-observer variation of the mean score of "erosions", "ulcerations" and endoscopic activity index in mild disease, and the scoring of "erythema" and "oedema" in moderate-severe disease was significantly higher in the inexperienced group.

A correlation was seen between all the observed endoscopic features in both groups of endoscopists. Among experienced endoscopists, a set of four endoscopic variables ("Vascular pattern", "Erosions", "Ulcerations" and Friability") explained 92% of the variation in EAI. By including "Granularity" in these set 91% of the variation in EAI was explained in the group of inexperienced endoscopists.

Conclusion

The inter-observer agreement in the rating of endoscopic features characterising ulcerative colitis is satisfactory in both groups of endoscopists but significantly higher in the experienced group. The difference in the mean score between the two groups is only significant for "ulcerations". The endoscopic variables "Vascular pattern", "Erosions", "Ulcerations" and Friability" explained the overall endoscopic activity index. Even though the present result is quite satisfactory, there is a potential of improvement. Improved grading systems might contribute to improve the consistency of endoscopic descriptions.