Medical image segmentation using deep learning: A survey

Abstract

Deep learning has been widely used for medical image segmentation and a large number of papers has been presented recording the success of deep learning in the field. A comprehensive thematic survey on medical image segmentation using deep learning techniques is presented. This paper makes two original contributions. Firstly, compared to traditional surveys that directly divide literatures of deep learning on medical image segmentation into many groups and introduce literatures in detail for each group, we classify currently popular literatures according to a multi‐level structure from coarse to fine. Secondly, this paper focuses on supervised and weakly supervised learning approaches, without including unsupervised approaches since they have been introduced in many old surveys and they are not popular currently. For supervised learning approaches, we analyse literatures in three aspects: the selection of backbone networks, the design of network blocks, and the improvement of loss functions. For weakly supervised learning approaches, we investigate literature according to data augmentation, transfer learning, and interactive segmentation, separately. Compared to existing surveys, this survey classifies the literatures very differently from before and is more convenient for readers to understand the relevant rationale and will guide them to think of appropriate improvements in medical image segmentation based on deep learning approaches.

Weakly supervised segmentation of COVID-19 infection with local lesion coherence on CT images

“…Medical image segmentation has been studied in the field of COVID-19 so far, and although many effective methods have been proposed, some problems have been found to be in urgent need of solution during the research process, and solving such problems is an excellent contribution to enhance the research results in this field. Despite the fact that the global outbreak region is extensive, data collection and tagging are difficult to complete in a short period of time due to high labor expenses and time constraints [11] , [38] . Weakly supervised learning is of particular research interest that are less labor and time compared to traditional training labeling methods.…”

“…Despite the fact that the global outbreak region is extensive, data collection and tagging are difficult to complete in a short period of time due to high labor expenses and time constraints [11] , [38] . Weakly supervised learning is of particular research interest that are less labor and time compared to traditional training labeling methods.…”

Auto-segmentation for total marrow irradiation

“…Due to their limited size, a very small deviation can lead to a very low Dice score. A known limitation of multi-layer deep learning in image recognition and segmentation is limited number of features ( 43 – 45 ) which may explain why optic nerves, and chiasm are among the worst scoring structures in Dice in the current study. However, average Dice 0.42-0.45 in the optic structures are not significantly lower than those reported in other studies, 0.37-0.65 ( 46 ) and 0.45-0.69 ( 47 ).…”

SM-SegNet: A Lightweight Squeeze M-SegNet for Tissue Segmentation in Brain MRI Scans

“…Zhou et al [ 34 ] proposed the U-net++ architecture, which has encoder and decoder blocks coupled with a number of layers and dense skip routes for medical image segmentation. A drawback in U-net++ is to significantly increase the number of parameters by using dense connections [ 35 ]. Gu et al [ 36 ] proposed the CE-Net, a context encoder network (CE-Net) that leverages a pre-trained ResNet block within the encoder to aid the segmentation of medical images.…”