OGC P14 A quantitative method to enhance endoscopic measurement

Abstract Background Upper gastrointestinal white light endoscopy is the most used procedure for assessing upper tract condition and identification of various types lesions. The characteristics lesions may influence subsequent treatment steps. Among these characteristics, size a lesion plays crucial role in determining surveillance interval, appropriate resection method predicting risk malignancy. Visual estimation highly subjective although this currently clinical practice. Another important characteristic location, which estimated using markers on scope. However, stomach endoscope be curved to visualise lesion, making location measurements inaccurate. Identifying accurate surgical planning help deciding extent required. We aimed improve endoscopic by developing quantitative that can provide an accurate, objective, contactless automatic way during procedure. Methods A miniaturised electromagnetic tracking sensor was places inside instrument channel trace pose camera An algorithm based feature points identify reference each image interest; triangulation technique then determine its precise with respect desired anatomical landmark. Surgeons only require take two different pictures one single picture any landmark they are interested knowing information. At same time corresponding information were recorded captured image. Using images (i.e. position orientation), our proposed computed distance selected tested simulated plastic model artificial lesion. Results 10 sets processed tract. first as soon had passed gastroesophageal junction; additional from angles when scope close rounded shape (diameter 12mm) placed 15.2 cm junction. method, junction 15.17 ± 0.51 (mean SD); assessed 12.76 0.83 mm. difference between mean actual values 2.28 mm 0.76 size. root square error 5.14mm (distance) 1.12 (size). Conclusions have developed validated accuracy model. This potentially better than deep learning approaches huge amount training data set. Our also compute directly whereas algorithms classify into groups. There no limitation tool independent. configuration accurately localised, clinically important. magnitude acceptable has potential efficiency diagnosis enhance patient outcome.