RESUMEN
Accurate and reliable segmentation of Gross Target Volume (GTV) is critical in cancer Radiation Therapy (RT) planning, but manual delineation is time-consuming and subject to inter-observer variations. Recently, deep learning methods have achieved remarkable success in medical image segmentation. However, due to the low image contrast and extreme pixel imbalance between GTV and adjacent tissues, most existing methods usually obtained limited performance on automatic GTV segmentation. In this paper, we propose a Heterogeneous Cascade Framework (HCF) from a decoupling perspective, which decomposes the GTV segmentation into independent recognition and segmentation subtasks. The former aims to screen out the abnormal slices containing GTV, while the latter performs pixel-wise segmentation of these slices. With the decoupled two-stage framework, we can efficiently filter normal slices to reduce false positives. To further improve the segmentation performance, we design a multi-level Spatial Alignment Network (SANet) based on the feature pyramid structure, which introduces a spatial alignment module into the decoder to compensate for the information loss caused by downsampling. Moreover, we propose a Combined Regularization (CR) loss and Balance-Sampling Strategy (BSS) to alleviate the pixel imbalance problem and improve network convergence. Extensive experiments on two public datasets of StructSeg2019 challenge demonstrate that our method outperforms state-of-the-art methods, especially with significant advantages in reducing false positives and accurately segmenting small objects. The code is available at https://github.com/shijun18/GTV_AutoSeg.
RESUMEN
Recently, convolutional neural network(CNN) has achieved great success in medical image segmentation. However, due to the limitation of convolutional receptive field, the pure convolutional neural network is difficult to further improve its performance. Given the outstanding ability of transformers in extracting the long-range dependency, some works have successfully applied it to computer vision and achieved better results than CNN in some tasks. Based on transformers could remedy the shortage of CNN, in this paper, we propose ITUnet, a segmentation network using CNN and transformers as features extractor. The combination of CNN and transformers enables the network to learn both short- and long-range dependency of features, which is beneficial to segmentation tasks. We evaluate our method on a head-and-neck CT dataset which has 18 kinds of organs to be segmented. The experimental results demonstrate that our proposed method shows better accuracy and robustness, the proposed methods achieve the Dice score of 77.72 and the 95% Hausdorff Distance of 2.31, outperforming the existing methods.
Asunto(s)
Procesamiento de Imagen Asistido por Computador , Órganos en Riesgo , Procesamiento de Imagen Asistido por Computador/métodos , Redes Neurales de la ComputaciónRESUMEN
Cervical cancer has become one of the important factors threatening women's health. Histopathological diagnosis is the most important criterion for cervical cancer diagnosis and treatment. Accurate classification of lesion degree of cervical epithelium by analyzing whole slide images (WSIs) can effectively improve the therapeutic effect and prognosis. However, classification of cervical lesion degree shows poor reproductivity due to lack of standardisation and is subjective among clinicians. In addition, due to the lack of large-scale finely annotated datasets, current deep learning methods do not perform well on this task. In this paper, we propose a two-stage method based on unsupervised pre-training to solve this multi-classification task. Our method first applied a patch-level network to predict the patch-level score and generate a heatmap that can highlight the lesion area. This network is pre-trained using an unsupervised method and verified on a public dataset. Then without extracting manual features, heatmaps are fed into a convolutional neural network (CNN) model directly for the WSI-level prediction. Our approach achieved an accuracy of 81.19% and a custom metric score of 0.9495 on the public cervical cancer WSI dataset, which is the highest in the public so far.
Asunto(s)
Neoplasias del Cuello Uterino , Femenino , Humanos , Redes Neurales de la Computación , Neoplasias del Cuello Uterino/diagnósticoRESUMEN
The use of total marrow and lymphoid irradiation (TMLI) as part of conditioning regimens for bone marrow transplantation is trending due to its advantages in disease control and low toxicity. Accurate contouring of target structures such as bone and lymph nodes plays an important role in irradiation planning. However, this process is often time-consuming and prone to inter-observer variation. Recently, deep learning methods such as convolutional neural networks (CNNs) and vision transformers have achieved tremendous success in medical image segmentation, therefore enabling fast semiautomatic radiotherapy planning. In this paper, we propose a dual-encoder U-shaped model named DE-Net, to automatically segment the target structures for TMLI. To enhance the learned features, the encoder of DE-Net is composed of parallel CNNs and vision transformers, which can model both local and global contexts. The multi-level features from the two branches are progressively fused by intermediate modules, therefore effectively preserving low-level details. Our experiments demonstrate that the proposed method achieves state-of-the-art results and a significant improvement in lymph node segmentation compared with existing methods.
Asunto(s)
Médula Ósea , Irradiación Linfática , Médula Ósea/diagnóstico por imagen , Trasplante de Médula Ósea , Redes Neurales de la ComputaciónRESUMEN
Precise segmentation of organs at risk (OARs) in computed tomography (CT) images is an essential step for lung cancer radiotherapy. However, the manual delineation of OARs is time-consuming and subject to inter-observer variation. Although U-like architecture has achieved great success in medical image segmentation recently, it exhibits the limitations in modeling long-range dependencies. As an alternative structure, Transformers have emerged due to the outstanding capability of capturing the global contextual information provided by Self-Attention(SA) mechanism. However, Transformers need more computational cost than CNNs for introducing the SA module. In this paper, we propose a novel module named fine-grained combination of Transformer and CNN(FcTC). FcTC module is composed of dual-path extractor and fusing unit to effectively extract local information and model long-distance dependency. Then we build FcTC-UNet to automatically segment the OARs in thoracic CT images. The experiments results demonstrate that the proposed method achieves better performance over other state-of-the-art methods.