We use cookies to understand how you use our site and to improve your experience. This includes personalizing content and advertising. To learn more, click here. By continuing to use our site, you accept our use of cookies. Cookie Policy.

MedImaging

Download Mobile App
Recent News Radiography MRI Ultrasound Nuclear Medicine General/Advanced Imaging Imaging IT Industry News

Groundbreaking AI Model Accurately Segments Liver Tumors from CT Scans

By MedImaging International staff writers
Posted on 02 Sep 2025

Liver cancer is the sixth most common cancer worldwide and a leading cause of cancer-related deaths. Accurate segmentation of liver tumors is critical for diagnosis and therapy, but manual methods by radiologists are time-consuming and prone to variations in expertise. Artificial intelligence (AI)-based segmentation has advanced the field, yet most models require thousands of cases for training, creating a barrier to adoption. Now, a groundbreaking AI model has shown high accuracy even with limited training datasets.

Researchers at the Institute of Science Tokyo (Tokyo, Japan) have developed the Multi-scale Hessian-enhanced Patch-based Neural Network (MHP-Net), a deep learning model for liver tumor segmentation. Instead of analyzing entire scans at once, MHP-Net divides images into 3D patches and pairs them with enhanced versions using Hessian filtering, which highlights spherical features like tumors. This architecture creates precise segmentation maps from contrast-enhanced CT scans.


Image: Patch-based deep-learning model with limited training dataset for liver tumor segmentation in contrast-enhanced hepatic CT (Yang et al. (2025), IEEE Access, 10.1109/ACCESS.2025.3570728)
Image: Patch-based deep-learning model with limited training dataset for liver tumor segmentation in contrast-enhanced hepatic CT (Yang et al. (2025), IEEE Access, 10.1109/ACCESS.2025.3570728)

The model’s accuracy was evaluated using Dice similarity scores, which measure overlap between predicted and expert-annotated tumor boundaries. Despite training on datasets as small as 7, 14, and 28 tumors, MHP-Net achieved Dice scores of 0.691, 0.709, and 0.719, respectively. The findings, published in IEEE Access, show the model outperformed established AI systems such as U-Net, Res U-Net, and HDense-U-Net.

In addition to its accuracy, MHP-Net’s lightweight design allows training in under 10 minutes and real-time inference in about four seconds per patient, making it suitable for clinical use with limited computing power. By reducing data requirements, it opens AI adoption to under-resourced hospitals worldwide. Researchers plan to expand the model’s applications, exploring small-data AI solutions for rare cancers and broader medical imaging challenges.

“This is just a start in the field of small-data AI, where meaningful and clinically relevant deep learning models can be built from limited datasets,” said Professor Kenji Suzuki, who led the development of the model. “MHP-Net’s success can inspire small-data AI solutions in other areas of medical imaging as well, such as the detection of rare cancers.”

Related Links:
Institute of Science Tokyo


High-Precision QA Tool
DEXA Phantom
Pocket Fetal Doppler
CONTEC10C/CL
New
Floor‑Mounted Digital X‑Ray System
MasteRad MX30+
Digital X-Ray Detector Panel
Acuity G4

Latest General/Advanced Imaging News

New CT-Based Indicator Helps Predict Life-Threatening Postpartum Bleeding Cases
02 Sep 2025  |   General/Advanced Imaging

CT Colonography Beats Stool DNA Testing for Colon Cancer Screening
02 Sep 2025  |   General/Advanced Imaging

First-Of-Its-Kind Wearable Device Offers Revolutionary Alternative to CT Scans
02 Sep 2025  |   General/Advanced Imaging