hemorrhage Microwave

The HFSS simulation system was used to construct a 3D model of human brain hemorrhage.

The 3D model is mainly divided into plasma, skin, skull, and tissue modules, and eight antenna arrays are set up at the periphery of the model. The innermost tissue module is set up as a sphere with a radius of 66mm. To mimic the real-world scenario of a human brain hemorrhage and enrich the diversity of the dataset, the center coordinates of the hemorrhage region, with x, y, and z values ranging from -30 to 30, were randomly generated. Additionally, the radius of the bleeding, varying between 0 and 9 mm, was randomly determined to ensure that the area remained confined within the tissue module. Microwave simulation experiments were carried out through the HFSS simulation system for bleed cases with different coordinates and radii. The resulting microwave experimental data were exported as the input data for the model. The x-y plane screenshot of the corresponding 3D model was used as the label data for the model, as shown in Fig. 4. The input data is a set of simulation data obtained from the antenna array at zero and thirty degrees each, and the subtraction of the data in the corresponding cells of the two sets of simulation data can effectively offset the noise and amplify the difference of the feature data so that the model can carry out effective learning and prevent it from being misled by the noise data. As can be seen in Fig. 4, the label data are colored, with both the human brain tissue part and the hemorrhagic mass part. The human brain tissue part is, in effect, constant in size and orientation, and only the orientation and size of the hemorrhagic mass are changing. The label data annotation tool is used to annotate the label images, converting them into greyscale images while discarding the color images. This process excludes noise in the label data and identifies the unchanged parts in the 3D model. Using the greyscale images as the label data allows the model to be fitted to only a single channel of the data images, reducing the number of features the model needs to accommodate. The greyscale image corresponding to an in Fig. 4 is as b is shown. The dataset used for the experiment contains 624 sets of data, divided into a training dataset and a test dataset; the training dataset uses 594 sets of data, and the test dataset contains 30 sets of data.