Computer Vision News - May 2016

Challenge Gland Segmentation Every month, Computer Vision News reviews a challenge related to our field, be it in medical imaging, automotive, robotics or else. If you don’t find time to read challenges, but are interested in the new methods proposed by the scientific community to solve them, you can read our challenge summaries! This month we have chosen to review the Gland Segmentation challenge , held in conjunction with the 18th International Conference on Medical Image Computing and Computer Assisted Interventions ( MICCAI 2015 in Munich, Germany ). The website of the challenge is here . COMPUTER VISION NEWS 21 Background Glands are important histological structures which are present in most organ systems and have the function of secreting chemicals called hormones to help the body function properly. Studies have shown that the most prevalent form of cancer is originated by adenocarcinomas, malignant tumors forming in mucus-secreting glands throughout the body. Adenocarcinomas are most prevalent in the following cancer types: prostate, pancreas, breast, oesophagus, lung and colon. The morphology of glands is closely examined by pathologists to assess the degree of malignancy of these tumors. Accurate segmentation of glands provides precious clues to morphological statistics. However, the majority of studies have focused until now on gland segmentation in healthy or benign samples, but rarely on intermediate or high grade cancer. The high degree of variation in glandular morphology in different histologic grades explains the need for run a gland segmentation study on a variety of histologic grades. The challenge requests participants to suggest a procedure and runs its algorithms on a high variability dataset consisting of 165 images featuring histological sections of stage T3 or T4 colorectal adenocarcinoma in as many different patients. Colorectal adenocarcinoma is the most common form of colon cancer and the morphology of intestinal glands is a crucial step to assess grade of the tumor, prognosis and treatment plan of individual patients. Microscopic image of glandular tissue Top Ranking Method The top ranking participant team suggested a novel deep contour-aware network exploring the multilevel feature representations with fully convolutional networks. The two parts of the procedure total 5 max-pooling layers and 3 up-sampling layers, where the former were used in the down- sampling path and the latter in the up- sampling path, to increase the resolutions of feature maps and to output the prediction masks.