An Across Feature Map Attention-Based Deep Learning Method for Small Liver Tumor Segmentation in CT Scans

Abstract

<h3>Purpose/Objective(s)</h3> Liver cancer is the second most frequent cancer to cause death in men and sixth for women. Detection of liver cancer early using computed tomography (CT) could assist doctors in making accurate hepatocellular carcinoma evaluation and treatment planning, which prevent millions of patients' death every year. Traditionally, radiologists delineate the liver lesion by manually reading hundreds of CT scans slice by slice, which is an enormous burden and suffers from inter- or intra-operator variations. Therefore, automatic liver tumor segmentation methods are highly demanded in clinical practice. Although many deep convolutional neural network-based techniques for automatic segmentation have been proposed, the small tumor remains challenging to segment due to convolutional and pooling operations that result in the loss of small object information. This study aims to propose a deep learning method for accurate segmentation for small liver tumors. <h3>Materials/Methods</h3> This study included 131 patients with liver cancer. The in-plane resolution of the patients' CTs is from 0.55 mm to 1.0 mm and slice spacing from 0.45 mm to 6.0 mm. We randomly selected 100 CT scans as the training set and others as the testing set. Each CT slice of the testing set was separated into groups according to tumor size as follows: 0.1–2.0, 2.1–5.0, 5.1–10.0, and 10.1–20.0 cm. The CT slice without tumor or tumor size > 20 cm were excluded. This work presents a novel attention-based deep learning method for small liver tumor segmentation, which exploits the relation between small and large tumors by computing the original image of small tumors and the feature maps of large tumors. The relation of small-large tumors can compensate for the information loss of small tumors during the convolutional and pooling operations and improve the performance of small tumor segmentation. <h3>Results</h3> Among 20,693 CT slices of the 31 testing patients, 3.0% CT slices with tumors ≤2 cm, 6.7% ≤5 cm, 10.6% ≤10 cm, and 13.4%≤20 cm. We compared our method with six widely used segmentation models. The results show our model outperforms other methods on all sizes of liver tumors, especially for small size tumors: For the 0.1-2.0 cm liver tumor, it achieved 8.4%, 10.0%, 11.3%, 9.1%, 10.9%, and 9.6% improvement compared to Unet, PAN, DeepLabV3, FPN, LinkNet, and PSPNet, respectively. <h3>Conclusion</h3> The tumors of different sizes share similar imaging characteristics. The information of large tumors can compensate for the information loss of small tumors from the feature propagation processes of deep learning methods. The small-large tumors relation can significantly improve small liver tumor segmentation, which is beneficial for disease diagnosis and treatment planning and has the potential for other types of tumor detection and segmentation.