Site Editor
Sandy Srinivas, MD
Sandy Srinivas, MD
Posted: Friday, May 26, 2023
A new treatment response assessment algorithm based in deep learning has been published in the journal Cancer Imaging. “Manual segmentation and measurement of diffusion-weighted imaging lesions based on Metastasis Reporting and Data System for Prostate Cancer [MET-RADS-P] require a high level of expertise, are time-consuming, and are subject to operator error,” said Xiaoying Wang, PhD, of Peking University First Hospital, China, and colleagues. “Deep learning technologies have extended this quantitative approach with promising preliminary results in the assessment of tumor response in the liver.”
The proof-of-concept study included 162 patients with advanced prostate cancer who had undergone treatment of metastasis. Each patient underwent at minimum of two scans for pelvic lymph nodes. The authors used a previous deep learning model for automated segmentation of patient lymph nodes. The evaluation pipeline included Dice similarity coefficient, volumetric similarity, and manual Bland-Altman plotting. Following lymph node analysis, the treatment response was automated via a rule-based algorithm that followed the MET-RADS-P criteria.
The authors reported that the mean Dice similarity coefficient and volumetric similarity values of patient pelvic lymph nodes were 0.82 ± 0.09 and 0.88 ± 0.12, respectively. Furthermore, Bland-Altman plotting demonstrated that most lymph node measurements were within the agreement limits. Automated segmentation was most accurate for the target lesions at 92% and 75% accurate for the nonpathologic lesions demonstrating sufficient sensitivity to detect pathologic lesions. Treatment response based on the automation of segmentation was reported by the group to be excellent for target lesions with a K value of 0.92 (0.86–0.98).
“The deep learning–based semiautomated algorithm showed high accuracy for the treatment response assessment of pelvic lymph nodes and demonstrated comparable performance with radiologists,” the investigators concluded.
Disclosure: The study authors reported no conflicts of interest.
JNCCN Spotlights
