Cystic kidney masses turn up all the time on CT scans taken for other reasons. Doctors sort them with the Bosniak scale, but lesions in its middle grades are hard to call, and the paper's authors say these masses are frequently overtreated: some people have surgery or ablation for a lesion that turns out to be benign.
What AI does
A team at Changhai Hospital, Naval Medical University, in Shanghai built KITE-CRM, a deep-learning pathway that reads one phase of a contrast CT scan, finds kidney abnormalities, decides which lesions are cystic and estimates how likely each is to be cancer. It shows outlines and a risk score inside the radiologist's normal reading software. Uncertain cases are flagged for team review, and radiologists can accept, edit or override every output. In the trial they fully agreed with the AI in 65.9% of exams and overrode it in 5.7%.
From November 2025 to February 2026, 5,869 consecutive adults having abdominal CT were randomly assigned to AI-assisted or standard reporting. Among lesions that went on to treatment, 28 of 171 (16.4%) were benign on pathology in the AI arm, against 104 of 258 (40.3%) with standard care. Across everyone randomized, that is about 26 fewer treatments of benign lesions per 1,000 people. Median reporting time fell from 18.5 to 7.6 minutes. Before the trial, six radiologists reading 117 hard cases with the AI scored higher accuracy and read faster, with the largest gains for the less experienced readers.
Hooray
Fewer people were treated for a kidney lesion that turned out to be harmless, and radiologists finished their reports in well under half the time. The team has posted the model code publicly so other hospitals can test it.
What comes next
The authors call for broader prospective trials, including in underserved populations, and longer follow-up on delayed diagnoses, cancer outcomes, kidney function and quality of life. They advise hospitals without kidney imaging specialists to treat a high AI score as a reason to refer patients to a specialist team, not as a reason for surgery.