Accuracy of an AI-developed questionnaire for early lung cancer screening.

Abstract

Since lung cancer (LC) symptoms are nonspecific and overlap with common respiratory diseases, early diagnosis of LC remains challenging. Therefore, we tasked artificial intelligence (AI, GPT-4o) with developing a questionnaire for the early detection of LC, incorporating symptoms and risk factors (RFs). Considering the significance of RFs, AI calculated the risk level in the questionnaire based on the total score: high risk (>30 points), moderate risk (16-30 points), and low risk (≤15 points). To validate the questionnaire, 100 individuals were surveyed (average age 64±10.9 years), including 60 females (60%) and 40 males (40%). 29 persons (29%) had confirmed LC. According to the survey results, 21 patients (21%) was considered to be at a high risk, including 18 of them (85.7%) with a confirmed LC (p<0.005). 52 patients could be classified as moderate risk group (52%), 9 (17.3%) from them had confirmed LC (p<0.005). 27 patients (27%) were considered to be at a low risk, and just 2 of them (7.4%) had confirmed LC (p<0.005). The AI-developed questionnaire demonstrated statistically significant accuracy in identifying high-risk patients, indicating its strong sensitivity. Patients at moderate risk require further evaluation using additional diagnostic methods, which is the focus of our future research.

Ähnliche Arbeiten