Lubomir M. Hadjiiski PhD, Presenter: Nothing to Disclose

Berkman Sahiner PhD, Abstract Co-Author: Nothing to Disclose

Heang-Ping Chan PhD, Abstract Co-Author: Nothing to Disclose

Naama R. Bogot MD, Abstract Co-Author: Nothing to Disclose

Philip Neil Cascade MD, Abstract Co-Author: Nothing to Disclose

Ella Annabelle Kazerooni MD, Abstract Co-Author: Nothing to Disclose

Ted Win Way MS, Abstract Co-Author: Nothing to Disclose

et al, Abstract Co-Author: Nothing to Disclose

Interval change analysis is an important technique used by radiologists to identify malignancy. We are developing a computer-aided diagnosis method to assist radiologists in differentiation of malignant and benign lung nodules by interval change analysis on serial CT scans.

An automated method was developed to extract and analyze features from corresponding lung nodules on temporal pairs of CT scans. Regions of interest containing the corresponding nodules were identified on the current and prior CT scans in the temporal pair. The lung nodules were automatically segmented using a constrained active contour model. Run length statistics (RLS) and spiculation features, as well as the volume were extracted from each nodule. Interval change features were calculated as the difference of the corresponding features extracted from the prior and the current scans of the same nodule. Stepwise feature selection with simplex optimization was used to select the best feature subset. A linear discriminant classifier was used to merge the selected features for classification of malignant and benign nodules.In this preliminary study, 41 temporal CT scans containing nodules were obtained from patient files. All malignant and most benign nodules were biopsy-proven and the remaining benign nodules were identified by 2-year follow-up. The true nodule locations were marked by an experienced radiologist. A leave-one-case-out resampling scheme was used for feature selection and classification. The classification accuracy was analyzed by the area (Az) under ROC curve.

An average of 2 features was selected from the training subsets. The most frequently selected features included a difference RLS feature and a volume feature. The classifier achieved a training Az of 0.91±0.05 and a test Az of 0.90±0.05. For comparison, the classifier using features extracted from the current CT scans alone achieved a training Az of 0.85±0.10 and a test Az of 0.79±0.08.

The difference RLS and volume features are useful for computerized classification of lung nodules on serial CT scans. Further studies are underway to improve the technique and to evaluate the performance on a larger data set.

Hadjiiski, L, Sahiner, B, Chan, H, Bogot, N, Cascade, P, Kazerooni, E, Way, T, et al, , Computer-aided Diagnosis of Lung Cancer: Interval Change Analysis of Nodule Features in Serial CT Examinations.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4416802.html