A predicting model of bone marrow malignant infiltration in 18F-FDG PET/CT images with increased diffuse bone marrow FDG uptake

Abstract

Purpose: To demonstrate the relationship between the etiologies of increased diffuse bone marrow (BM) 18F-FDG uptake and PET/CT imaging/clinical features, as well as to explore a predicting model of BM malignant infiltration (MI) based on decision tree. Methods: 84 patients with increased diffuse BM uptake were retrospectively enrolled. Their complete case record and PET/CT images were reviewed, with the maximal standardized uptake values of bone marrow (SUVmaxBM) and other imaging/clinical features were noted. At the same time, the differences in imaging/clinical features between bone marrow MI and non-MI groups were compared. The decision tree for predicting MI was established by C5.0 component of SPSS Clementine. Results: In patients with homogenously increased BM uptake, 21 patients had MI resulted from leukemia, lymphoma and small cell lung cancer (SCLC). MI group had higher SUVmaxBM than non-MI group (6.7±3.1 vs 4.2±0.9, p=0.001). However, a considerable proportion of MI patients had similar SUVmaxBM to non-MI patients, which were mainly seen in lymphoplasmacytic lymphoma/Waldenström macroglobulinemia (LPL/WM), chronic myeloid leukemia (CML) and multiple myeloma (MM). There were significant differences in other factors between the two groups. MI patients were highly associated with SUVmaxAP/AX≥1 (the ratio of SUVmaxBM of appendicular skeleton to that of axial skeleton), hepatosplenomegaly, older age and lower rate of fever. The decision tree combining SUVmaxBM, SUVmaxAP/AX, fever and hepatosplenomegaly achieved a sensitivity of 81.0%, a specificity of 98.4% and an accuracy of 94.0% for predicting MI. Conclusion: Increased diffuse BM 18F-FDG uptake can be attributed to both bone marrow MI and benign etiologies. A decision tree based on C5.0 algorithm, combining PET/CT imaging and clinical features, is of potential use in discriminating BM malignant infiltration from patients with increased diffuse BM uptake.

Keywords: PET/CT; SUVmax; bone marrow; malignant infiltration; predicting model.

Conflict of interest statement

Competing Interests: The authors have declared that no competing interest exists.

Figures

Figure 1

Increased diffuse BM uptake on PET images. (A) A 49-year-old woman was diagnosed with lymphoblastic leukemia from bone marrow biopsy. Both axial and appendicular skeleton had diffuse FDG accumulation. The SUVmax of bone marrow was 9.1, which was higher than the cutoff value (6.15). (B) A 41-year-old woman was diagnosed with bone marrow hyperplasia by bone marrow biopsy. Diffuse FDG accumulation was distributed mainly in axial skeleton. The SUVmax of bone marrow was 4.5, which was lower than the cutoff value (6.15).

Figure 2

PET images in a 66-year-old woman diagnosed with LPL/WM before and after therapy. (A) PET image before chemotherapy showed mildly increased bone marrow uptake both in axial and appendicular skeleton, with a SUVmax of 3.7. (B) No increased bone marrow uptake was seen on PET image after chemotherapy.

Figure 3

Comparison of SUVmaxBM in different groups. A, non-MI group (N=63). B, MI group (N=21). C, Patients diagnosed with LPL/WM, multiple myeloma and CML in MI group (N=9). D. Other patients in MI group (N=12).

Figure 4

A model for predicting bone marrow malignant infiltration based on decision tree. It was executed by C5.0 component of SPSS Clementine.