Lung and Lung Tumor Segmentation of CT Images During MWA Therapy Using AI Algorithm

Abstract

Microwave ablation (MWA) therapy as a thermal ablation procedure is an excellent alternative to open surgery for tumor treatment. The technique is considered advantageous for patients who are not candidates for open surgery due to factors such as age, anatomic limitations, resection, etc. Computed tomography (CT) is a commonly used interventional imaging modality during MWA therapy for localizing the tumor and finalizing the tumor treatment process. However, the CT scan of the body usually includes neighboring organs that are not relevant to lung tumor MWA therapy. Therefore, the segmentation of the lung and lung tumor in CT images provides valuable information about the tumor margin. This information can assist physicians in precisely and completely destroying the tumor during the MWA procedure. To solve the aforementioned problem, deep learning (DL), in particular, achieves a higher level of accuracy in segmentation than machine learning techniques due to its composition of multiple learning layers. The immediate goal is to distinguish among the different tissue structures of the tumor, healthy tissue, and the ablated area in lung CT images using the DL method to segment the organ and cancer area. Researchers have proposed various segmentation models. However, different segmentation tasks require different perception fields. In this study, we propose a new DL model that includes a residual block based on the U-Net model to accurately segment the lung organ and lung tumor tissue. The dataset consists of lung CT images acquired during MWA therapy using a CT scanner at the University Hospital Magdeburg. Manual tumor segmentation has been performed and confirmed by physicians. The results of our proposed method can be compared with those of the U-net model with a SSIM of 90%. Furthermore, accurately determining the margin area of the tumor tissue can decrease insufficient tumor ablation, which often leads to tumor recurrence. We anticipate that our proposed model can be generalized to perform tumor segmentation on CT images of different organs during MWA treatment. Finally, we hope that this method can achieve sufficient accuracy to decrease tumor recurrence and enable dose reduction for patients in interventional CT imaging. Doi: 10.28991/SciMedJ-2023-05-01-01 Full Text: PDF