TOWARDS AN UNIFIED FRAMEWORK FOR VALIDATING DEEP LEARNING METHODS FOR COLORECTAL POLYPS: FIRST STEPS

INTRODUCTION

Deep learning techniques have significantly contributed to the field of medical imaging analysis. In case of colorectal cancer, they have shown a great utility for increasing the adenoma detection rate at colonoscopy, but a common validation methodology is still missing. In this study, we present preliminary efforts towards the definition of a validation framework.

MATERIAL AND METHODS

Different models based on different backbones and encoder-decoder architectures have been trained with a publicly available dataset that contains white light and NBI colonoscopy videos, with 76 different lesions from colonoscopy procedures in 48 human patients. A computer aided detection (CADe) demonstrator has been implemented to show the performance of the models.

RESULTS

This CADe demonstrator shows the areas detected as polyp by overlapping the predicted mask on the endoscopic image. It allows selecting the video to be used, among those from the test set. Although it only present basic features such as play, pause and moving to the next video, it easily loads the model and allows for visualization of results. The demonstrator is accompanied by a set of metrics to be used depending on the aimed task: polyp detection, localization and segmentation.

CONCLUSIONS

The use of this CADe demonstrator, together with a publicly available dataset and predefined metrics will allow for an easier and more fair comparison of methods. Further work is still required to validate the proposed framework.

A markerless motion capture system to study musculoskeletal biomechanics: visual hull and simulated annealing approach

Human motion capture is frequently used to study musculoskeletal biomechanics and clinical problems, as well as to provide realistic animation for the entertainment industry. The most popular technique for human motion capture uses markers placed on the skin, despite some important drawbacks including the impediment to the motion by the presence of skin markers and relative movement between the skin where the markers are placed and the underlying bone. The latter makes it difficult to estimate the motion of the underlying bone, which is the variable of interest for biomechanical and clinical applications. A model-based markerless motion capture system is presented in this study, which does not require the placement of any markers on the subject's body. The described method is based on visual hull reconstruction and an a priori model of the subject. A custom version of adapted fast simulated annealing has been developed to match the model to the visual hull. The tracking capability and a quantitative validation of the method were evaluated in a virtual environment for a complete gait cycle. The obtained mean errors, for an entire gait cycle, for knee and hip flexion are respectively 1.5 degrees (+/-3.9 degrees ) and 2.0 degrees (+/-3.0 degrees ), while for knee and hip adduction they are respectively 2.0 degrees (+/-2.3 degrees ) and 1.1 degrees (+/-1.7 degrees ). Results for the ankle and shoulder joints are also presented. Experimental results captured in a gait laboratory with a real subject are also shown to demonstrate the effectiveness and potential of the presented method in a clinical environment.