The author of this document has limited its availability to on-campus or logged-in CSUSB users only.

Off-campus CSUSB users: To download restricted items, please log in to our proxy server with your MyCoyote username and password.

Date of Award


Document Type

Restricted Project: Campus only access

Degree Name

Master of Science in Computer Science


School of Computer Science and Engineering

First Reader/Committee Chair

Jin, Jennifer


Liver disease is a significant global health concern, necessitating the development of advanced diagnostic tools for accurate and timely diagnosis. Liver segmentation and lesion detection from medical images play an essential role in assessing liver disease, enabling physicians to evaluate liver volume, morphology, and the presence of lesions. The objective of this study is to develop an automatic liver segmentation and lesion detection algorithm using a deep learning-based U-Net model, which can improve the diagnostic accuracy and efficiency for various liver diseases.

In this study, we propose a novel approach to automatically segment liver tissues and detect liver lesions in medical images using a U-Net model, a popular deep learning architecture for image segmentation tasks. The U-Net model comprises an encoder-decoder structure that captures context and localization information to generate accurate segmentation masks. Our model is trained and tested on a dataset consisting of liver and lesion images obtained from multiple imaging modalities, including magnetic resonance imaging (MRI) and computed tomography (CT) scans.

To improve the segmentation performance, we use a combined loss function that incorporates both the Dice coefficient loss and binary cross-entropy loss. This combination effectively balances the trade-off between the accurate delineation of liver boundaries and maintaining the overall structure of liver tissue. Furthermore, we employ various data preprocessing techniques, such as image resizing and normalization, to enhance the model's generalizability and robustness against different imaging modalities and liver conditions.

Our model's performance is assessed using several evaluation metrics, including Dice coefficient, sensitivity, specificity, Jaccard index, precision, and F1-score. These metrics provide a comprehensive understanding of the model's segmentation accuracy and its ability to discriminate between liver tissue, lesion regions, and background. Additionally, we extract lesion regions from the segmented liver masks using a threshold-based approach and analyze the lesion characteristics, such as size and shape, to provide valuable insights into the liver's condition.

The results of our study demonstrate that the proposed deep learning-based U-Net model can achieve accurate and efficient liver segmentation and lesion detection in medical images. Our model outperforms conventional image processing techniques and other deep learning models in terms of segmentation accuracy and computational efficiency. Moreover, the extracted lesion information can potentially aid in the early diagnosis of liver diseases, monitoring of disease progression, and planning of surgical interventions.

In conclusion, our study presents a novel and effective approach to liver segmentation and lesion detection in medical images using a deep learning-based U-Net model. The proposed method has the potential to improve the diagnosis and management of liver diseases by providing accurate and reliable liver segmentation and lesion detection. Furthermore, our approach can be extended to other organs and tissues, demonstrating its applicability and versatility in various medical image analysis tasks.