Within the validation studies, PIQE underwent performance (bench) testing using computational analysis as well as a multi-center, randomized, blinded clinical image review.
Initial performance testing, based on images of the American College of Radiology (medium sized) resolution phantom, was performed using metrics of signal-to-noise ratio (SNR) and signal intensity profiles for ringing and sharpness. Results indicated that PIQE can generate higher in-plane matrix from lower matrix images and contributes to ringing artifact reduction and increase of sharpness. The performance testing also included evaluation in typical clinical images of brain and knee, based on the calculated metrics of Edge Slope Width (to evaluate image sharpness), Ringing Variable Mean (to evaluate ringing artifacts), SNR, and Contrast Change Ratio. This further bench testing confirmed that PIQE (as compared to other typical techniques) generates images with sharper edges while mitigating the smoothing and ringing effects and maintaining similar or better contrast and SNR, even when applied to tripling the matrix size in both dimensions (i.e., an overall factor of 9x).
Because PIQE is a fundamentally new technique, and because it relies on machine learning based algorithms, it was important to evaluate the performance more rigorously with a human observer study to quantify image quality as well as diagnostic quality. Within the clinical image review, images were anonymized, randomized, blinded (i.e., stripped of identifying information like reconstruction techniques), and distributed for detailed image evaluation and scoring to six US board-certified radiologists, three specializing in neuro imaging, and three in musculoskeletal imaging. The images were reconstructed with either the conventional method or PIQE at various matrix sizes and for various scan durations. For the image evaluation, characteristics such as ringing artifact, image sharpness, SNR, overall image quality (IQ) and feature conspicuity were scored by the radiologists using a modified 5-point Likert scale, where a score of three (3) or above is considered clinically acceptable. A total of 53 unique subjects, from multiple sites in USA, Europe, and Japan, were scanned at either 3T (Vantage Galan 3T) or 1.5T (Vantage Orian or Vantage Fortian) in brain or knee to provide the data sets comprising a total of 544 clinical scans, representing multiple orientations (axial, sagittal and coronal), and multiple contrast weightings (T1-/T2-/PD-weighted with/without Fat saturation) within the FSE2D family of pulse sequences. In particular, based on the claims of increasing matrix size, reducing scan time, and mitigating ringing, various comparisons between low matrix acquisition with high matrix reconstruction versus high matrix acquisition and standard filters were performed. The clinical image review, segregated and performed by field strength, resulted in all images scoring at, or above, the clinically acceptable level by the Radiologists. The reviewers exhibited a strong agreement at the “good” and “very good” level for all IQ metrics such as SNR, image sharpness, image ringing, overall IQ and feature conspicuity. Furthermore, the scores were statistically compared across the various combinations of acquired vs reconstructed matrix sizes (and PIQE vs traditional filters) to identify similarities and superiority.