AI Can Help Radiologists Avoid Missed Lung Cancer on Chest X-Ray Images

Chest radiography is the most frequently performed first-line imaging examination, often used for routine health checkups, preoperative evaluations, and lung cancer screenings in primary healthcare settings worldwide. However, false-negative or misleading chest radiography results are common, and unexpected abnormalities like lung cancer may be overlooked. Missing lung cancer can result in the disease progressing from an early-stage, potentially curable state to advanced-stage lung cancer, leading to poor patient outcomes. Now, a recent analysis has identified common features of missed lung cancer cases on chest X-ray images, suggesting that computer-aided detection systems or artificial intelligence could assist radiologists in avoiding such errors.

Researchers at Ramathibodi Hospital (Bangkok, Thailand) analyzed chest X-rays from 75 patients at a single institution, where initial signs of lung cancer were overlooked. They found common features of missed lung cancer cases, including the tumor's location, density, and any superimposed anatomical structures. According to the researchers, radiologists can prevent such errors by comparing the chest radiograph being read with previous examples, focusing on common blind spots, avoiding misidentification of lesions as normal structures, and using computer-aided detection systems or artificial intelligence.

Image: A new study has found some common features of missed lung cancer cases on chest X-ray images (Photo courtesy of Freepik)

The study retrospectively reviewed chest X-rays of 95 patients taken at least six months before a lung cancer diagnosis. The final sample consisted of 75 patients, with an even gender distribution and an average age of nearly 65 years. The median missed lung cancer size was approximately 16 mm. Most cases (75%) occurred in the outer two-thirds of the lung, followed by 68% in the middle/lower zones and 55% in the left lung. Common features included anatomical superimposition (88% of cases), density lower than the aortic knob (85%), partly or poorly defined margin (77%), irregular or spiculated border (61%), and rounded or oval shape (57%). Almost 47% of the missed cases had stage 3 or 4 lung cancer, and 41% of patients died.

The findings indicate that certain radiographic features of missed lung cancer, such as density, location, and superimposed anatomical structures, identified in the first positive chest radiograph were linked to advanced-stage lung cancer diagnosis and higher 3-year all-cause mortality. Delayed detection of missed lung cancer with such radiographic features on chest radiography may lead to worse outcomes. Hence, radiologists should pay special attention to and promptly report equivocal or borderline lesions in the upper zone or inner one-third of the lung, those with density equal to or greater than the aortic knob, or those superimposed by midline structures, pulmonary vessels, and ribs found incidentally on chest radiographs. The researchers hope that these findings will help radiologists avoid errors that can result in poor patient outcomes and death. Furthermore, computer-aided detection systems or artificial intelligence could improve radiologist performance and help detect smaller tumors, suggested the researchers.

Related Links:
Ramathibodi Hospital