Image analysis: 68Ga-FAPI-46 PET derived texture parameters improve the differentiation of malignant and benign pulmonary lesions

Abstract

Background: Pulmonary lesions inconclusive in ¹⁸F-FDG PET/CT are a known clinical problem. Both texture analysis and ⁶⁸Ga-FAPI-46 have shown potential in thoracic oncological problems but their combination has not been assessed yet. This initial analysis aims to evaluate the utility of ⁶⁸Ga-FAPI-46 PET texture parameters to differentiate between lung cancer and benign pulmonary lesions inconclusive in ¹⁸F-FDG PET/CT.

Materials and methods: 20 histologically confirmed pulmonary lesions (13 lung cancer, 7 benign) in 19 patients were evaluated. All patients underwent an inconclusive ¹⁸F-FDG PET/CT before ⁶⁸Ga-FAPI-46 PET/CT. 64 texture parameters and conventional parameters (SUVs, TBRs) were analyzed. Texture parameters with significant (P < 0.05) differences between lung cancer and benign lesions were detected by the Mann-Whitney U test. Boxplots and a scatter plot matrix were created. Principal component analyses and Spearman correlations were performed. Receiver operating characteristics curves with area under the curve (AUC) values were created for univariable and bivariable logistic regression.

Results: The texture parameters HIST Maximum grey level (AUC = 0.901), HIST Mean (AUC = 0.802), HIST Mode (AUC = 0.835), HIST Range (AUC = 0.901) and GLCM Information correlation 1 (AUC = 0.824) showed significant differences between lung cancer and benign pulmonary lesions. AUC values of conventional parameters (SUVmax, SUVmean, TBR(SUVmax), TBR(SUVmean)) were 0.791, 0.868, 0.802 and 0.857, respectively. Maximum AUC values of bivariable logistic regression were 0.967 and 0.978 for two texture parameters and the combination of conventional and texture parameters, respectively. Correlations between texture parameter pairs were mainly moderate (0.4≤ρ≤0.59). 2/5 texture parameters (HIST Mean, HIST Mode) displayed no very strong correlations (0.8≤ρ≤1.00) to any conventional parameters or lesion volume.

Conclusion: ⁶⁸Ga-FAPI-46 PET texture parameters show great potential to differentiate between lung cancer and benign pulmonary lesions inconclusive in ¹⁸F-FDG/PET. Spearman correlations indicate additional information value of texture parameters.

Keywords: Image analysis; Lung cancer; Pulmonary lesion; Texture analysis.

Fig. 1

Methodic overview

Fig. 2

Boxplots of conventional (A) and texture parameters (B). Boxes represent the interquartile range with the horizontal line inside each box indicating the median. Whiskers extend to 1.5 time the interquartile range. Individual data points are shown as dots. Statistical significance is marked with stars (*=p < 0.05, **=p < 0.01)

Fig. 3

Scatter plot matrix of texture parameters with Spearman correlations and histograms. Spearman correlation coefficients between two different texture parameters are given in the upper right section of the diagram for all data, only lung cancer data and only data of benign lesions, respectively. Statistical significance of correlation is marked with stars (* = p < 0.05, ** = p < 0.01, *** = p < 0.001). Scatter plots in the lower left section visually show distribution of texture parameter values. The diagonal displays the histograms of individual texture parameters. Red dots and boxes represent lung cancer data while green dots and boxes represent data of benign lesions

Fig. 4

Spearman matrix displaying correlation coefficients for the relationships between texture parameters and conventional parameters

Fig. 5

Principal component analysis of all texture parameters (A) and only the statistically significant texture parameters (B)

Fig. 6

ROC curves with AUC values of univariable logistic regression of conventional parameters (A) and texture parameters (B)

Fig. 7

ROC curves with AUC values of bivariable logistic regression of statistically significant texture parameter pairs (A) and combinations of conventional parameters with statistically significant texture parameters (B)