RESUMO
PURPOSE: To conduct a simplified lesion-detection task of a low-dose (LD) PET-CT protocol for frequent lung screening using 30% of the effective PETCT dose and to investigate the feasibility of increasing clinical value of low-statistics scans using machine learning. METHODS: We acquired 33 SD PET images, of which 13 had actual LD (ALD) PET, and simulated LD (SLD) PET images at seven different count levels from the SD PET scans. We employed image quality transfer (IQT), a machine learning algorithm that performs patch-regression to map parameters from low-quality to high-quality images. At each count level, patches extracted from 23 pairs of SD/SLD PET images were used to train three IQT models - global linear, single tree, and random forest regressions with cubic patch sizes of 3 and 5 voxels. The models were then used to estimate SD images from LD images at each count level for 10 unseen subjects. Lesion-detection task was carried out on matched lesion-present and lesion-absent images. RESULTS: LD PET-CT protocol yielded lesion detectability with sensitivity of 0.98 and specificity of 1. Random forest algorithm with cubic patch size of 5 allowed further 11.7% reduction in the effective PETCT dose without compromising lesion detectability, but underestimated SUV by 30%. CONCLUSION: LD PET-CT protocol was validated for lesion detection using ALD PET scans. Substantial image quality improvement or additional dose reduction while preserving clinical values can be achieved using machine learning methods though SUV quantification may be biased and adjustment of our research protocol is required for clinical use.
Assuntos
Neoplasias Pulmonares , Tomografia por Emissão de Pósitrons combinada à Tomografia Computadorizada , Algoritmos , Humanos , Neoplasias Pulmonares/diagnóstico por imagem , Aprendizado de Máquina , Tomografia por Emissão de PósitronsRESUMO
OBJECTIVE: To compare commercially available image analysis tools Hermes BRASS and Siemens Syngo.VIA with clinical assessment in 18F-Florbetapir PET scans. METHODS: 225 scans were reported by clinicians and quantified using two software packages. Scans were classified into Type A (typical features) or non-Type A (atypical features) for both positive and negative scans. For BRASS, scans with z-score ≥ 2 in 2 ≥ region of interest were classed positive. For Syngo.VIA a positive scan was indicated when mean cortical standardized uptake value ratio (mcSUVR) ≥ 1.17. RESULTS: 81% scans were Type A, and 19% scans were non-Type A. The sensitivity of BRASS and Syngo.VIA for Type A scans was 98.8 and 96.3%, specificity was 73 and 92%, respectively. Sensitivity for non-Type A scans was 95.8 and 79.2%, specificity was 36.8 and 57.9%, respectively.A third threshold of identifiable levels of plaque (1.08 ≤ mcSUVR ≤ 1.17) was recommended for Syngo.VIA to increase detection of false negative scans.The false positive rate of BRASS significantly decreased when an alternative positive threshold value of mcSUVR ≥ 1.18.Introduction of alternative criteria did not improve prediction outcome for non-Type A scans. More complex solutions are recommended. CONCLUSION: Hermes criteria for a positive scan leads to a high sensitivity but a low specificity. Siemens Syngo.VIA criteria gives a high sensitivity and specificity and agrees better with the clinical report. Alternative thresholds and classifications may help to improve agreement with the clinical report. ADVANCES IN KNOWLEDGE: Software packages may assist with clinical reporting of more difficult to interpret cases that require a more experienced read.
Assuntos
Peptídeos beta-Amiloides/metabolismo , Encéfalo/diagnóstico por imagem , Demência/diagnóstico por imagem , Interpretação de Imagem Assistida por Computador/métodos , Neuroimagem/métodos , Tomografia por Emissão de Pósitrons/métodos , Adulto , Idoso , Idoso de 80 Anos ou mais , Encéfalo/metabolismo , Encéfalo/patologia , Demência/metabolismo , Demência/patologia , Estudos de Avaliação como Assunto , Feminino , Humanos , Masculino , Pessoa de Meia-Idade , Sensibilidade e EspecificidadeRESUMO
OBJECTIVES: This study investigates the usefulness of quantitative SUVR thresholds on sub types of typical (type A) and atypical (non-type A) positive (Aß+) and negative (Aß-) 18F-florbetapir scans and aims to optimise the thresholds. METHODS: Clinical 18F-florbetapir scans (n = 100) were categorised by sub type and visual reads were performed independently by three trained readers. Inter-reader agreement and reader-to-reference agreement were measured. Optimal SUVR thresholds were derived by ROC analysis and were compared with thresholds derived from a healthy control group and values from published literature. RESULTS: Sub type division of 18F-florbetapir PET scans improves accuracy and agreement of visual reads for type A: accuracy 90%, 96% and 70% and agreement κ > 0.7, κ ≥ 0.85 and -0.1 < κ < 0.9 for all data, type A and non-type A respectively. Sub type division also improves quantitative classification accuracy of type A: optimum mcSUVR thresholds were found to be 1.32, 1.18 and 1.48 with accuracy 86%, 92% and 76% for all data, type A and non-type A respectively. CONCLUSIONS: Aß+/Aß- mcSUVR threshold of 1.18 is suitable for classification of type A studies (sensitivity = 97%, specificity = 88%). Region-wise SUVR thresholds may improve classification accuracy in non-type A studies. Amyloid PET scans should be divided by sub type before quantification. ADVANCES IN KNOWLEDGE: We have derived and validated mcSUVR thresholds for Aß+/Aß- 18F-florbetapir studies. This work demonstrates that division into sub types improves reader accuracy and agreement and quantification accuracy in scans with typical presentation and highlights the atypical presentations not suited to global SUVR quantification.