Back to the list
Congress: ECR25
Poster Number: C-27006
Type: Poster: EPOS Radiologist (scientific)
Authorblock: M. Balaguer-Montero1, A. Marcos Morales1, M. Ligero2, D. Leiva1, L. M. Atlagich1, N. Staikoglou1, C. Zatse1, C. Monreal1, R. Perez Lopez1; 1Barcelona/ES, 2Dresden/DE
Disclosures:
Maria Balaguer-Montero: Nothing to disclose
Adrià Marcos Morales: Nothing to disclose
Marta Ligero: Nothing to disclose
David Leiva: Nothing to disclose
Luz Maria Atlagich: Nothing to disclose
Nikolaos Staikoglou: Nothing to disclose
Christina Zatse: Nothing to disclose
Camilo Monreal: Nothing to disclose
Raquel Perez Lopez: Nothing to disclose
Keywords: Artificial Intelligence, Liver, Oncology, CT, Computer Applications-General, Observer performance, Segmentation, Cancer
Purpose Liver tumors, whether primary or metastatic, significantly impact cancer patients' outcomes. Accurate identification and precise quantification of liver cancer are crucial for effective patient management, including precise diagnosis, prognosis, and evaluation of anticancer therapy efficacy. The evaluation of liver tumor burden is crucial at different stages of cancer treatment, and it is typically performed on medical images, such as computed tomography (CT). Currently, this task is carried out manually by radiologists, which is not only time-consuming but also prone to...
Read more Methods and materials Study populationThe model was developed and validated on 1598 contrast-enhanced CT scans of the liver, coming from 1306 patients with cancer (20% of them without liver tumor burden), accounting for a total of 4908 liver tumors. The images used for the development and testing of the model came from an internal dataset comprised of 885 CT scans from 593 patients, whereas the external validation data was gathered from four independent open-access repositories [2-5] (Figure 1). [fig 1] The whole dataset comprises a wide...
Read more Results Accuracy of SALSAThe nnU-Net framework clearly outperformed the other two architectures in segmentation capabilities (Figure 3), making it the representative model for SALSA. [fig 3] Our proposed model shows high accuracy in liver tumor detection with a patient-wise precision of 99.65%, and a recall of 94.17% for the external validation cohort. When considering each lesion individually, SALSA obtained a lesion-by-lesion detection precision of 81.72% and a recall of 57.92% in the same dataset (Figure 4, Table 2). [fig 4] [fig 2] In parallel, the tumor masks automatically generated...
Read more Conclusion SALSA achieves precise and fully automated identification and delineation of liver cancer on CT images, facilitating more accurate quantification of tumor burden, a critical factor in cancer prognosis and management. The model achieved superior accuracy in tumor identification and volume quantification, outperforming state-of-the-art tools and exceeding the inter-reader agreement among three expert radiologists. SALSA holds promise for improving cancer care, enhancing patient outcomes, and increasing healthcare system efficiency by providing rapid, consistent, and reliable data on liver cancer count and...
Read more References Eisenhauer, E.A., Therasse, P., Bogaerts, J., Schwartz, L.H., Sargent, D., Ford, R., Dancey, J., Arbuck, S., Gwyther, S., Mooney, M., et al. (2009). New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur. J. Cancer 45, 228–247. https://doi.org10.1016/j.ejca.2008.10.026 Bilic, P., Christ, P., Li, H.B., Vorontsov, E., Ben-Cohen, A., Kaissis, G., Szeskin, A., Jacobs, C., Mamani, G.E.H., Chartrand, G., et al. (2023). The Liver Tumor Segmentation Benchmark (LiTS). Med. Image Anal. 84, 102680. https://doi.org/10.1016/j.media.2022.102680 Antonelli, M., Reinke, A., Bakas, S.,...
Read more
GALLERY