AI-Based Automated Lipomatous Tumor Segmentation in MR Images: Ensemble Solution to Heterogeneous Data.

Deep learning (DL) has been proposed to automate image segmentation and provide accuracy, consistency, and efficiency. Accurate segmentation of lipomatous tumors (LTs) is critical for correct tumor radiomics analysis and localization. The major challenge of this task is data heterogeneity, including tumor morphological characteristics and multicenter scanning protocols. To mitigate the issue, we aimed to develop a DL-based Super Learner (SL) ensemble framework with different data correction and normalization methods. Pathologically proven LTs on pre-operative T1-weighted/proton-density MR images of 185 patients were manually segmented. The LTs were categorized by tumor locations as distal upper limb (DUL), distal lower limb (DLL), proximal upper limb (PUL), proximal lower limb (PLL), or Trunk (T) and grouped by 80%/9%/11% for training, validation and testing. Six configurations of correction/normalization were applied to data for fivefold-cross-validation trainings, resulting in 30 base learners (BLs). A SL was obtained from the BLs by optimizing SL weights. The performance was evaluated by dice-similarity-coefficient (DSC), sensitivity, specificity, and Hausdorff distance (HD95). For predictions of the BLs, the average DSC, sensitivity, and specificity from the testing data were 0.72 [Formula: see text] 0.16, 0.73 [Formula: see text] 0.168, and 0.99 [Formula: see text] 0.012, respectively, while for SL predictions were 0.80 [Formula: see text] 0.184, 0.78 [Formula: see text] 0.193, and 1.00 [Formula: see text] 0.010. The average HD95 of the BLs were 11.5 (DUL), 23.2 (DLL), 25.9 (PUL), 32.1 (PLL), and 47.9 (T) mm, whereas of SL were 1.7, 8.4, 15.9, 2.2, and 36.6 mm, respectively. The proposed method could improve the segmentation accuracy and mitigate the performance instability and data heterogeneity aiding the differential diagnosis of LTs in real clinical situations.

Qualitative evaluation of MRI features of lipoma and atypical lipomatous tumor: results from a multicenter study.

OBJECTIVES: The objectives of the study are (1) to distinguish lipoma (L) from atypical lipomatous tumor (ALT) using MRI qualitative features, (2) to assess the value of contrast enhancement, and (3) to evaluate the reproducibility and confidence level of radiological readings. MATERIALS AND METHODS: Patients with pathologically proven L or ALT, who underwent MRI within 3 months from surgical excision were included in this retrospective multicenter international study. Two radiologists independently reviewed MRI centrally. Impressions were recorded as L or ALT. A third radiologist was consulted for discordant readings. The two radiologists re-read all non-contrast sequences; impression was recorded; then post-contrast images were reviewed and any changes were recorded. RESULTS: A total of 246 patients (135 females; median age, 59 years) were included. ALT was histopathologically confirmed in 70/246 patients. In multivariable analysis, in addition to the lesion size, deep location, proximal lower limb lesions, demonstrating incomplete fat suppression, or increased architectural complexity were the independent predictive features of ALT; but not the contrast enhancement. Post-contrast MRI changed the impression in a total of 5 studies (3 for R1 and 4 for R2; 2 studies are common); all of them were incorrectly changed from Ls to ALTs. Overall, inter-reader kappa agreement was 0.42 (95% CI 0.39-0.56). Discordance between the two readers was statistically significant for both pathologically proven L (p < 0.001) and ALT (p = 0.003). CONCLUSION: Most qualitative MR imaging features can help distinguish ALTs from BLs. However, contrast enhancement may be limited and occasionally misleading. Substantial discordance on MRI readings exists between radiologists with a relatively high false positive and negative rates.