Automated measurement of detectability index in CT imaging: Development and validation

The purpose of this study is to develop software for measuring the detectability index (d') automatically from ACR 464 CT phantom images.

Software for measuring d' automatically was developed with Python 3.9.13 using the PyQt5 graphical user interface (GUI) as part of the IndoQCT platform. The task-transfer function (TTF) and noise power spectrum (NPS) were automatically measured to obtain spatial resolution and noise texture information. The task function was defined with a Gaussian and flat types, a matrix size of 300 pixels, a pixel size of 0.05 mm, and a contrast of 15 HU. The task object diameter was set to 5 mm for the tube current and kernel type variations, and ranged from 1 to 15 mm for the object diameter variation. The task object contrast ranged from 1-29 HU for the object contrast variation. Each dataset was evaluated in terms of the d' using the non-pre-whitening (NPW) model observer. Images of an ACR 464 CT phantom scanned using a GE Revolution EVO scanner with tube currents of 80, 100, 120, 140, 160, and 200 mA and kernel types of Standard, Edge, Lung, and Soft were used for evaluation. The results of our developed software were compared with ImQuest results.

In general, our developed software produced d' values that were in strong agreement with ImQuest across all tested variations and both task function types (Gaussian and flat). For tube currents, an increase in tube current consistently increased the d' value (r = 0.98). Analysis of kernel types showed that the Standard kernel yielded the highest detectability, while the Lung kernel yielded the lowest. For variations in task object diameter and contrast, larger diameters and higher contrasts increased detectability following exponential (R² > 0.99) and linear trends (R² = 1), respectively. Across all variations (kernel, object diameter, and contrast), the correlation between IndoQCT and ImQuest was exceptionally strong (r > 0.98), validating the performance of the developed software.

Software to automatically measure the d' has been successfully developed. It is easily accessible with a straightforward, fast, accurate, and intuitive workflow.