Channelized hotelling and human observer correlation for lesion detection in hepatic SPECT imaging.

Area of Science:

• Medical Imaging
• Computational Modeling
• Observer Performance Studies

Background:

• Mathematical model observers are valuable for predicting human performance in medical imaging, potentially replacing human observers in psychophysical studies.
• The channelized Hotelling observer (CHO) is a type of model observer of interest for these applications.

Purpose of the Study:

• To examine the correlations between human observers and several forms of the channelized Hotelling observer (CHO).
• To evaluate the effectiveness of scatter correction strategies in single-photon-emission computed tomography (SPECT) liver imaging for tumor detection.

Main Methods:

• A receiver operating characteristic (ROC) study was conducted using simulated SPECT liver images with varying scatter conditions and correction strategies.
• Human observer performance was assessed by averaging areas under the ROC curve for 7 individuals across different tumor locations and scatter handling strategies.
• Signal-to-noise ratios (SNRs) from human performance were compared with SNRs from constant-Q and difference-of-Gaussian (DOG) CHO models.

Main Results:

• A strong Spearman rank correlation (0.92, P = 0.000001) was observed between average human performance and a constant-Q CHO model.
• A significant correlation (0.84, P = 0.00005) was also found between human performance and the DOG version of the CHO.

Conclusions:

• The significant positive correlations indicate that channelized model observers can potentially replace human observers in certain medical imaging tasks.
• The studied CHO models are effective for preliminary screening of different imaging strategies before conducting human psychophysical studies.