Automated Liver Tumor Segmentation and Detection for Enhanced Diagnosis

Automated Liver Tumor Segmentation and Detection for Enhanced Diagnosis

Karthik Kurupapu

Student,

Dept. of Computer Science, GITAM Deemed To Be University, Visakhapatnam, India

Email: Kurupapukarthik@gmail.com

Mrs. K. Yasudha

Assistant Professor, Dept. of Computer Science

GITAM Deemed To Be University, Visakhapatnam, India Email:ykomali@gitam.edu

Abstract—Liver tumor segmentation is critical for accurate diagnosis and treatment planning. This system employs deep learning models such as ResNet-50 (main algorithm), DeepLabv3+ with ResNet-101, VGG-16, and Fully Convolutional Networks (FCNs) to segment liver and tumor regions from medical images. ResNet-50 is chosen for its efficiency and strong feature extraction. DeepLabv3+ captures multi-scale features, VGG-16 offers a strong baseline, and FCNs provide dense pixel-level predictions. These models improve segmentation accuracy and reduce reliance on manual annotation. Evaluation is performed using metrics like Dice Score and IoU. The approach aims to support radiologists with precise, automated tumor localization. This enhances early detection and improves clinical outcomes.

**Index Terms—**ResNet-50, DeepLabv3+, FCN, VGG-16, Liver Tumor Segmentation, Medical Imaging.