Accuracy and reproducibility of adipose tissue measurements in young infants by whole body magnetic resonance imaging

Abstract

Purpose: MR might be well suited to obtain reproducible and accurate measures of fat tissues in infants. This study evaluates MR-measurements of adipose tissue in young infants in vitro and in vivo.

Material and methods: MR images of ten phantoms simulating subcutaneous fat of an infant's torso were obtained using a 1.5T MR scanner with and without simulated breathing. Scans consisted of a cartesian water-suppression turbo spin echo (wsTSE) sequence, and a PROPELLER wsTSE sequence. Fat volume was quantified directly and by MR imaging using k-means clustering and threshold-based segmentation procedures to calculate accuracy in vitro. Whole body MR was obtained in sleeping young infants (average age 67±30 days). This study was approved by the local review board. All parents gave written informed consent. To obtain reproducibility in vivo, cartesian and PROPELLER wsTSE sequences were repeated in seven and four young infants, respectively. Overall, 21 repetitions were performed for the cartesian sequence and 13 repetitions for the PROPELLER sequence.

Results: In vitro accuracy errors depended on the chosen segmentation procedure, ranging from 5.4% to 76%, while the sequence showed no significant influence. Artificial breathing increased the minimal accuracy error to 9.1%. In vivo reproducibility errors for total fat volume of the sleeping infants ranged from 2.6% to 3.4%. Neither segmentation nor sequence significantly influenced reproducibility.

Conclusion: With both cartesian and PROPELLER sequences an accurate and reproducible measure of body fat was achieved. Adequate segmentation was mandatory for high accuracy.

Fig 1. The phantom scanned by MR imaging with the cartesian wsTSE sequence without (A) and with breathing simulation (B) and the PROPELLER wsTSE sequence without (C) and with breathing simulation (D).

Fig 2. Infant scans with the cartesian wsTSE sequence (A,B) and the PROPELLER wsTSE sequence (C,D) and the corresponding threshold-based segmentations using a threshold of 150.

Fig 3. Infant scans with the cartesian wsTSE sequence (A), the k-means clustering segmentation of subcutaneous fat (B), and one threshold-based segmentation of subcutaneous fat using a threshold of 150 (C).

The k-means clustering segmentation better matches the visual findings (inserts).

Fig 4. Five representative slices of one infant.

Left row: cartesian wsTSE sequence; middle row: original segmentation using the k-means clustering algorithm; right row: manually corrected segmentation with separation of internal (green) and external fat (red). This scan represents a case with insufficient water suppression at the arms, where the most user interaction among all scanned infants was required. The complete scan is available as (S1 Fig.).