A novel computer-aided diagnosis (CAD) algorithm has bested three un-aided thoracic radiologists at predicting malignancy in small lung nodules found on low-dose CT imaging, according to a study published online Aug. 5 in Radiology.
The CAD method had diagnostic accuracy of 91 percent versus 70 percent for the board-certified radiologists, who used up-to-date practice guidelines, and the improvement held even when patients were matched by numerous variables and risk factors.
Lead author Peng Huang, PhD, of Johns Hopkins Medicine and colleagues note that their algorithm analyzes not only surrounding lung tissue but also the texture of uncalcified nodules, which in the present study measured 20 millimeters or less in diameter.
Further, they report, CAD increased positive predictive value from 0.64 to 0.86 and decreased the false-positive rate from 0.31 to 0.11
The team selected a matched case-control sample of 186 patients with small lung nodules who underwent biopsy in the National Lung Screening Trial.
They randomly split the patients’ pre-biopsy CT exams into a training set (70 cancers plus 70 benign controls) and a validation set (20 cancers plus 26 benign controls).
Huang and colleagues describe their CAD algorithm as one that uses machine learning from analysis of a training set. They tested the algorithm in a validation set using prespecified cutoff values derived from the training set.
In their discussion, the authors note that radiologists typically risk stratify noncalcified indeterminate pulmonary nodules by interpreting nodule characteristics such as diameter, volume, margin, attenuation and location.
However, “none of these variables alone is sufficient to accurately classify indeterminate pulmonary nodules, because of significant overlap among risk categories and complicated interactions among features,” they point out. “[M]any benign nodules, such as granulomas, could have a malignant appearance at CT because of spiculated or lobulated margins without a subsolid or ground-glass component.”
The researchers underscore that their CAD algorithm performed well even when analyzing imaging of lung nodules smaller than 10 millimeters.
They surmise that the superior diagnostic performance of CAD they observed probably owed, in no small part, to the higher prediction accuracy of the independent validation data from the CAD system as compared with the three radiologists’ readings.
In any case, their findings “suggest that CAD substantially reduces the low-dose CT screening false-positive rate and increases the positive predictive value of lung nodule evaluation,” they write.
“By helping radiologists distinguish benign lesions from malignant ones, CAD has the potential to reduce the morbidity associated with low-dose CT screening, including radiation exposure, overdiagnosis of incidental findings and anxiety, as well as to reduce unnecessary testing and the financial costs of lung cancer screening,” Huang et al write.
Enhanced accuracies ahead?
The authors acknowledge a number of limitations in their study design, including the selectivity and small size of their sample, as well as their use of only one dominant nodule per image for feature extraction.
The upside of the latter aspect was that it allowed the researchers to avoid being clued in by correlations among nodules in the same patient, the authors note.
“Several prediction models have recently been proposed to estimate the probability of cancer for detected nodules by using demographic risk factors and limited image features such as nodule size, volume, margin, attenuation and count,” the authors conclude. “Addition of nodule and non-nodule image texture features to these prediction models could substantially increase their accuracies for lung cancer screening.”