Fig. 2: Quantitative accuracy evaluated with local metrics on external scanners and external tracers. | Nature Communications

Fig. 2: Quantitative accuracy evaluated with local metrics on external scanners and external tracers.

From: Using domain knowledge for robust and generalizable deep learning-based CT-free PET attenuation and scatter correction

Fig. 2

Quantitative accuracy of the DL ASC-PET images generated with our proposed method, to the CT ASC-PET on the cross-scanner (a–d) and cross-tracer (e–h) setting, evaluated with clinical imaging parameters such as SUVmean, SUVmax, total lesion glycolysis (TLG), as well as the most relevant radiomics features. Data are summarized by a box and whisker plot (the central line in the box plot indicates the median of the data, while the edges of the box indicate the 25th and 75th percentiles; extending from the box are whiskers, the top whisker expands to the maxima and the bottom whisker to the minima). Sample size: a–d: Siemens Vision 450 (n = 51), Siemens Vision 600 (n = 62), UI uMI 780 (n = 98), GE Discovery MI (n = 104); e–h: 68Ga-FAPI (n = 7), 68Ga-DOTA-TATE (n = 17), 68Ga-DOTA-TOC (n = 8), 18F-PSMA (n = 12).

Back to article page