VISUAL ASSESSMENT OF β-AMYLOID PET SCAN IS IMPROVED BY CAPAIBL

Visual assessment is the regulatory authorities’ approved method for the interpretation of Aβ PET scans. A tracer-specific visual assessment has to be performed by trained readers. However, accurate visual interpretation of PET scans can be challenging, particularly for less experienced readers. In this study, we investigated the added value of using semiquantitative reports, generated using the image analysis software CapAIBL®, alongside visual Aβ image inspection. Forty-eight subjects with a wide spectrum of dementia diagnoses confirmed by autopsy, were identified from two independent databases. Subjects underwent either a 11C-PiB or 18F-florbetaben PET scan; Centiloid (CL) values were generated using CapAIBL. Four readers (CR, PR, TC and VD), blinded to clinicopathological diagnosis, visually categorized all scans in regards to Aβ plaque load (0 no Aβ load, 1 minor Aβ load, 2 significant Aβ load, 3 severe Aβ load). Grade 0 and 1 were classified as low Aβ (Aβ-), while grade 2 and 3 as high Aβ (Aβ+). Non-expert readers (PR TC and VD) did not follow tracer-specific training and were asked to rate confidence of their image interpretation. They were subsequently instructed to re-examine the scans alongside the CapAIBL reports, which included the CL values and surface projection of the Aβ load z-scores. Correlation of expert visual read with CapAIBL CL yielded optimal threshold values of 25.5CL between Aβ- and Aβ+ scans (sensitivity:1.00, specificity:0.95, accuracy:0.98, AUC:0.99) and 45CL between significant and severe Aβ load (sensitivity:1, specificity:1, accuracy:1, AUC:1). When compared to the expert, the performance of the non-expert group was, sensitivity:0.84, specificity: 0.98 acccuracy:90 and AUC: 0.91. On the second read, when provided with the CapAIBL reports, these statistics reached 0.96, 0.97, 0.97 and 0.97, respectively. On the first read, the intra-reader reliability (IRR) was (Fleiss's Kappa) 0.64 on the Aβ status and 0.41 on the Aβ load grade. The IRR reached 0.87 and 0.56 on the last read with CapAIBL reports. All non-expert readers were significantly more confident in their scan interpretation when using CapAIBL report (p<0.01). The addition of CapAIBL reports alongside to visual Aβ image inspection improved scan interpretation as well as inter-reader reliability among non-expert readers.