An artificial intelligence (AI)-based approach showed comparable diagnostic performance to experienced radiologists in detecting gallbladder cancer at a hospital in Chandigarh, according to a study published in The Lancet Regional Health – Southeast Asia .
Gallbladder cancer (GBC) is a highly aggressive malignancy with low detection rates and high mortality. Early diagnosis is challenging because benign gallbladder lesions may have similar imaging features, the researchers said.
A team from the Postgraduate Institute of Medical Education and Research (PGIMER), Chandigarh, and the Indian Institute of Technology (IIT), New Delhi aimed to develop and validate a deep learning (DL) model for GBC detection using abdominal ultrasound and compare its performance with radiologists.
Deep learning is a method in artificial intelligence that is inspired by the human brain and teaches computers to process data.
The study used abdominal ultrasound data of patients with gallbladder lesions obtained at PGIMER, a tertiary hospital, between August 2019 and June 2021.
The deep learning (DL) model was trained on a dataset of 233 patients, validated on 59 patients, and tested on 273 patients.
The performance of the DL model was evaluated in terms of sensitivity, specificity, and area under the receiver operating characteristic curve (AUC), which is widely used to measure the accuracy of diagnostic tests.
Two radiologists also independently reviewed the ultrasound images and compared their diagnostic performance with the DL model.
The study showed that in the test set, the DL model detected GBC with a sensitivity of 92.3%, a specificity of 74.4%, and an AUC of 0.887, which was comparable to that of two radiologists.
The researchers stated that the DL-based method showed high sensitivity and AUC in detecting GBCs in the presence of stones, gallbladder contraction, small lesions (less than 10 mm), and neck lesions, which was also consistent with radiologists’ The level is comparable.
They said the DL model showed greater sensitivity in detecting GBC wall thickening patterns compared with a single radiologist, albeit with reduced specificity.
“The deep learning-based approach demonstrated comparable diagnostic performance to experienced radiologists in detecting GBC using ultrasound,” the study’s authors noted.
They added: “Further multicenter studies are recommended to fully explore the potential of DL-based GBC diagnosis.”
The authors acknowledge some limitations of the study. The findings are based on a single-center data set and require multi-center studies for more extensive validation.
They added that the study’s knowledge cut-off date is 2021 and subsequent advances in DL and GBC diagnosis may not be reflected.
Svlook
