Release date: 2017-11-02

A malignant tumor that occurs in the mesenchymal tissue is called a "sarcoma." The word Sarcoma comes from the Greek word meaning "succulent growth." There are two main types of sarcoma: osteosarcoma and soft tissue sarcoma. Osteosarcoma is a kind of primary malignant bone tumor with great harm. The early onset symptoms are similar to other bone injuries. If it can not be found in time and treated accordingly, it is prone to lung metastasis. Once lung metastasis occurs, the hope of cure will be greatly increased. cut back.

At present, the diagnostic steps for suspected cases of osteosarcoma are physical examination → imaging examination → pathological examination. Computed tomography (CT) is one of the necessary means for the imaging examination of osteosarcoma, which can better show the boundary of cortical damage and three-dimensional anatomy. According to the CT image, the doctor can make a more reasonable radiotherapy or chemotherapy plan. However, manual delineation of the tumor area takes a long time and requires a lot of work. Moreover, the results of the delineation of the tumor area by different doctors are affected by many factors such as subjective experience and environment, and the reproducibility of the sketched results is poor. Therefore, there is a need to develop a method that can automatically segment a tumor area.

However, the automatic segmentation of CT images of osteosarcoma faces many challenges: (1) Osteosarcoma is not only located on the bone, but also invades the soft tissues around the bone and the muscle tissue at the joints. Therefore, the distribution of gray scale and texture in osteosarcoma is not uniform, and the difference in gray scale between tumor tissue and surrounding normal tissue is small, and the edge of the tumor is blurred (as shown in Fig. 1). (2) Different patients have different tumor morphological features and locations. (3) Imaging equipment and imaging parameters vary from hospital to hospital, resulting in differences in patient images in different hospitals.

Figure 1. CT image of osteosarcoma and tumor area boundaries (solid red line). The first line is the CT image of osteosarcoma, and the second line is the result of the border delineation of the tumor area. The three columns from left to right are tumors located in muscle tissue, bone and mixed lesions.

Recently, the Medical Imaging Room of the Suzhou Medical Institute of the Chinese Academy of Sciences proposed a method for automatically segmenting tumor regions, called Multiple Supervised Fully Convolutional Networks (MSFCN) based on multi-supervised full convolutional neural network. Using this method, researchers have trained a computer artificial intelligence assistant. This small assistant can automatically complete the tumor area delineation and reduce the burden on doctors by learning the method of delineating the tumor area boundary from experienced radiologists. At the same time, because the machine has no feelings, it will not be affected by mood or environmental factors, so the repeatability of the tumor area is improved.

This method sounds very long. In fact, there are two main points: first, multi-supervised; second, full convolutional neural network. The so-called "multi-supervised", the computer "learns" the radiologist's delineation method from multiple levels of detail. The "full-convolution neural network" means that the computer can accept images of any size as input and directly output the same size segmentation as the original image, without any post-processing steps.

The MSFCN method is based on the framework of a full convolutional neural network, adding a number of supervised edge output layers. Each edge output layer can calculate the loss value by the loss function by comparing the gold standard (standard answer), and then back propagation. The loss value information, in turn, guides the neural network to learn the multi-scale features, thereby obtaining the local features and global features of the image at the same time, and retains the related related information in the image more on the sampling of the network. Finally, the researchers used a fusion layer with weights to fuse the classification results to obtain the final tumor segmentation boundary results.

The experimental results show that compared with other advanced algorithms such as full convolutional neural network, this method has better adaptability and evaluation effect in many aspects such as similarity and sensitivity. The results of the study are published in the Journal of Computer Methods and Programs in Biomedicine (https://doi.org/10.1016/j.cmpb.2017.02.013).

Figure 2. CT image segmentation results for mixed osteosarcoma. Lines 1~4 are 4 different samples. (a) Gold standard (standard answer): Tumor area boundary (red line) sketched by experienced radiologists; (b) ~ (e) Segmentation results of FCN, U-Net, HED, MSFCN algorithms The boundaries of the tumor area are indicated by solid lines of different colors. As can be seen from the figure, the MSFCN method has a smoother boundary than the pattern split by other algorithms, and has the highest degree of agreement with the "standard answer".

In addition to assisting physicians in developing chemoradiotherapy or surgical procedures, the MSFCN method can also be used in accelerated imaging omics, imaging marker construction, and other medical imaging-assisted diagnostic assays (Figure 3).

Figure 3. Image omics analysis / imaging marker construction method flow: First, the researcher divides the tumor region into the tumor region to obtain the tumor region; then, extracts the image features such as gray scale and shape from the tumor region; The relationship between these image features and clinical data information is analyzed to construct a predictive model.

In addition, the medical imaging room of Suzhou Medical Institute has also studied the early prediction of tumor treatment effect, and proposed a DCE-MRI volume transfer constant (Ktrans) voxel analysis method.

The researchers used DCE-MRI image data of patients with soft tissue sarcoma before and after radiotherapy and chemotherapy, two weeks after radiotherapy and chemotherapy, and the tumor necrosis rate (TCNR) confirmed by surgical pathology after radiotherapy and chemotherapy. The researchers aligned the DCE-MRI images before and after radiotherapy and chemotherapy, and then based on the well-delineed tumor regions, constructed a Vtrans map of the voxel-based tumor region, and then calculated a significant increase in the tumor region Ktrans according to this graph. Volume fractions (F+, F-, and F0) with reduced and no significant changes were used as predictors of efficacy. Furthermore, the effect of early predictive measures of efficacy was evaluated (Fig. 4).

Figure 4. Voxel distribution of different efficacy cases. The therapeutic effect from a to d is weakened in turn. It can be seen visually that the more blue regions or the fewer green regions, the worse the treatment effect. From a to d, F0 is sequentially reduced (green area), F- is sequentially increased (blue area), TCNR is sequentially decreased, and the therapeutic effect is sequentially decreased.

The experimental results show that the volume prediction scores of F- and F0 generated by this method are better than the current traditional tumor evaluation indicators. In addition, the experimental results can be explained by the theory of tumor vascular normalization, which has good clinical significance. The relevant research results are published in the European Journal of Radiology (https://doi.org/10.1016/j.ejrad.2017.08.021).

Source: Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Sciences

Source: Voice of the Chinese Academy of Sciences (micro signal zkyzswx)

Fruit and Vegetable Powder

chlorella powder,spirulina powder, blue spirulina powder , rice protein etc

Organic Powder,Butterfly Pea Powder,Chaga Mushroom Extract,Hericium Erinaceus Extract

Youth Biotech CO,. Ltd. , https://www.youtherb.com