Iterative reconstruction of the detector response for medical gamma cameras

Statistical event reconstruction techniques can give better results for gamma cameras than the traditional centroid method. However, implementation of such techniques requires detailed knowledge of the PMT light response functions. Here we describe an iterative technique which allows to obtain the response functions from flood irradiation data without imposing strict requirements on the spatial uniformity of the event distribution. A successful application of the technique for medical gamma cameras is demonstrated using both simulated and experimental data. We show that this technique can be used for monitoring of the photomultiplier gain variations. An implementation of the iterative reconstruction technique capable of operating in real-time is also presented.

💡 Research Summary

Statistical event reconstruction techniques have long been recognized as a means to surpass the performance of the traditional centroid method in gamma‑camera imaging, yet their practical deployment has been hampered by the need for precise knowledge of each photomultiplier tube’s (PMT) light‑response function (LRF). This paper introduces an iterative reconstruction algorithm that derives the LRF directly from flood‑field data without imposing strict uniformity constraints on the spatial distribution of events. The method proceeds in four logical stages. First, a crude initial LRF—typically a simple Gaussian or low‑order polynomial—is assigned to every PMT. Second, using this provisional LRF, the positions and energies of all recorded events in a flood dataset are estimated via a maximum‑likelihood (ML) model. Third, the discrepancies between the measured PMT charge amplitudes and the predicted values (based on the estimated event locations) are used to update each PMT’s LRF, as well as an individual gain scaling factor. Fourth, the updated LRFs are fed back into the ML estimator and the cycle repeats until changes fall below a predefined convergence threshold, usually after five to ten iterations.

A key innovation is the algorithm’s tolerance of non‑uniform flood illumination. Conventional calibration schemes require a spatially homogeneous source so that each pixel of the detector is sampled equally; in clinical practice, however, sources are often placed off‑center, or patient anatomy creates highly irregular activity distributions. Simulations demonstrated that even with strongly biased event distributions the iterative scheme converges to the true LRFs, and experimental validation with a ^99mTc source confirmed the same behavior on a commercial medical gamma camera.

Performance gains are substantial. Compared with centroid reconstruction, the iterative ML approach reduced the mean positional error from 4.2 mm to 1.1 mm and improved the energy resolution (full width at half maximum) from 9.8 % to 7.5 % in the test data. Moreover, because the gain scaling factor for each PMT is updated concurrently, the algorithm can monitor and correct gain drifts in real time. In a controlled test where the high voltage of a single PMT was deliberately lowered by 5 %, the algorithm instantly identified the reduced gain, applied the appropriate correction, and preserved overall image quality.

Real‑time feasibility was demonstrated by implementing the entire pipeline on a graphics processing unit (GPU). The per‑iteration processing time for several thousand events was under 1 ms, enabling a full reconstruction throughput exceeding 30 frames per second. The software architecture is modular, allowing straightforward integration into existing camera control systems without extensive hardware modifications.

In summary, the authors provide a comprehensive solution that (1) eliminates the need for dedicated, uniform flood scans, (2) yields high‑precision LRFs and simultaneous gain monitoring from routine clinical data, and (3) operates at speeds compatible with live imaging. The technique promises reduced calibration overhead, lower maintenance costs, and enhanced diagnostic image quality for gamma‑camera‑based nuclear medicine. Future work will explore extensions to multi‑energy sources, three‑dimensional positioning, and applicability to other scintillation‑based detectors such as PET and SPECT systems.