Lung Tumor Image Segmentation from Computer Tomography Images Using MobileNetV2 and Transfer Learning

Background: Lung cancer is one of the most fatal cancers worldwide, and malignant tumors are characterized by the growth of abnormal cells in the tissues of lungs. Usually, symptoms of lung cancer do not appear until it is already at an advanced stage. The proper segmentation of cancerous lesions in CT images is the primary method of detection towards achieving a completely automated diagnostic system. Method: In this work, we developed an improved hybrid neural network via the fusion of two architectures, MobileNetV2 and UNET, for the semantic segmentation of malignant lung tumors from CT images. The transfer learning technique was employed and the pre-trained MobileNetV2 was utilized as an encoder of a conventional UNET model for feature extraction. The proposed network is an efficient segmentation approach that performs lightweight filtering to reduce computation and pointwise convolution for building more features. Skip connections were established with the Relu activation function for improving model convergence to connect the encoder layers of MobileNetv2 to decoder layers in UNET that allow the concatenation of feature maps with different resolutions from the encoder to decoder. Furthermore, the model was trained and fine-tuned on the training dataset acquired from the Medical Segmentation Decathlon (MSD) 2018 Challenge. Results: The proposed network was tested and evaluated on 25% of the dataset obtained from the MSD, and it achieved a dice score of 0.8793, recall of 0.8602 and precision of 0.93. It is pertinent to mention that our technique outperforms the current available networks, which have several phases of training and testing.

Automated summary

In this study, an improved hybrid neural network was developed by merging MobileNetV2 and UNet architectures for semantic segmentation of malignant lung tumors from CT images. Transfer learning was used, where the pre-trained MobileNetV2 served as an encoder for feature extraction in a conventional UNet model. Skip connections with Relu activation function were established to connect the encoder and decoder layers, enabling concatenation of feature maps with different resolutions. The network was trained and fine-tuned using the training dataset from the MSD 2018 Challenge. Testing on a subset of the MSD dataset resulted in a dice score of 0.8793, recall of 0.8602, and precision of 0.93. The technique outperforms current available networks with multiple training and testing phases.