Medical imaging, such as computed tomography (CT) scans, plays a crucial role in oncology, offering essential data for cancer detection, treatment planning, and monitoring of response to therapies. However, one of the key challenges in using these images effectively is the accurate delineation of tumors, a task known as tumor contouring. This process, which is vital for volume analysis, is often time-consuming and prone to significant variability between different observers. This variability can create bottlenecks in both clinical settings and research projects, especially when dealing with volumetric disease assessments. Primary liver cancers, including hepatocellular carcinoma (HCC), are frequently diagnosed at later stages when treatment options are limited, and the prognosis is often poor. Additionally, the liver is a common site for metastasis from other primary cancers, which complicates treatment and affects overall outcomes. In response to these challenges, researchers have developed an innovative deep learning-based tool to improve the diagnosis, treatment planning, and monitoring of liver cancer.

The new tool, called SALSA (System for Automatic Liver tumor Segmentation And detection), was developed by researchers at the Vall d'Hebron Institute of Oncology (VHIO, Barcelona, Spain; www.vhio.net) to automate the detection and tracking of liver tumors. This AI-powered system works directly with CT images, automatically detecting and delineating both primary and metastatic liver tumors. To build and train SALSA, the researchers used the nnU-Net segmentation framework, incorporating data from 1598 CT scans that included 4908 liver tumors, both primary and metastatic.

SALSA achieved exceptional results, demonstrating a tumor detection precision of over 99% at the patient level, and nearly 82% lesion-by-lesion detection precision in an external validation cohort. The integration of artificial intelligence and machine learning in this tool signifies a breakthrough in personalized oncology, offering a more accurate method for assessing treatment response on an individual basis. By outperforming existing state-of-the-art models and improving the inter-observer agreement among radiologists, SALSA shows promise in enhancing liver cancer detection, optimizing treatment strategies, and improving response evaluation.

“This novel deep learning-driven tool has shown precise and automated identification and delineation of liver cancer on CT images, facilitating a more precise quantification of tumor burden —a crucial factor in cancer prognosis and treatment—with no prior manual prompt requirements,” said Raquel Perez-Lopez, corresponding author of the paper published in Cell Reports Medicine. “Our validation across several test and external cohorts highlights SALSA’s effectiveness and reliability, matching, and often surpassing, the accuracy of expert radiologists.”

http://www.allmycbs.com/en/index.asp