Digital differentiation of non-small cell carcinomas of the lung by the fractal dimension of their epithelial architecture.

INTRODUCTION: In recent years, differences have emerged in the treatment of squamous and non-squamous non-small cell lung carcinomas (NSCLCs). This highlights the importance of accurate histopathologic classification. However, there remains inter-observer disagreement when making diagnoses based on histology. Fractal dimension (FD) is a mathematical measure of irregularity and complexity of shape. We hypothesize that the FD of carcinoma epithelial architecture can assist in differentiating adenocarcinoma (ADC) from squamous cell carcinoma (SCC) of the lung. METHODS: 134 resected (88 ADC and 46 SCC) cases of resected early-stage NSCLC were analyzed. Tissue micro arrays were generated from formalin-fixed paraffin-embedded tissue, stained with pan-cytokeratin, and digitally imaged and the FD of the epithelial structure calculated. Mean FD of ADC and SCC were compared using the independent t-test, partial correlations, and receiver operating characteristic (ROC) analyses. RESULTS: A statistically significant difference (p<0.001) between the mean FD of ADC (M=1.70, SD=0.07) and SCC (M=1.78, SD=0.07) was found. Significance remained (p<0.001) when controlling for several possible confounders. ROC analysis demonstrated an area-under-the-curve of 0.81 (p<0.001). CONCLUSIONS: The epithelial structure FD of NSCLC has potential as a reproducible and automated measure to help subtype NSCLCs into ADC and SCC. With further image analysis algorithm improvements, fractal analysis may be a component in computerized histomorphological assessments of lung cancer and may provide an adjunct test in differentiating NSCLCs.