Optimizing complex imaging procedures within Computed Tomography, considering both dose and image quality, presents significant challenges amidst rapid technological advancements and the adoption of machine learning (ML) methods. A crucial metric in this context is the Difference-Detailed Curve, which relies on human observer studies. However, these studies are labor-intensive and prone to both inter- and intra-observer variability. To tackle these issues, a ML-based model observer utilizing the U-Net architecture and a Bayesian methodology is proposed. In order to train a model observer unaffected by the spatial arrangement of low-contrast objects, the image preprocessing incorporates a Gaussian Process-based noise model. Additionally, gradient-weighted class activation mapping is utilized to gain insights into the model observer's decision-making process. By training on data from a diverse group of observers, well-calibrated probabilistic predictions that quantify observer variability are achieved. Leveraging the principles of Beta regression, the Bayesian methodology is used to derive a model observer performance metric, effectively gauging the model observer's strength in terms of an 'effective number of observers'. Ultimately, this framework enables to predict the DDC distribution by applying thresholds to the inferred probabilities (Part of this work has been presented at: Stocker D, Sommer C, Gueng S, Stäuble J, Özden I, Griessinger J, Weyland M S, Lutters G, Scheidegger S (2023). Probabilistic U-Net Model Observer for the DDC Method in CT Scan Protocol Optimization. The 56th SSRMP Annual Meeting 2023, November 30. - December 1., 2023, Luzern, Switzerland).
Image Processing, Computer-Assisted , Tomography, X-Ray Computed , Image Processing, Computer-Assisted/methods , Humans , Bayes Theorem , Machine Learning , Observer Variation
BACKGROUND: Quantifying femoral and tibial torsion is crucial in the preoperative planning for derotation surgery in children and adolescents. The use of an ultra-low-dose computed tomography (CT) protocol might be possible for modern CT scanners and suitable for reliable torsion measurements even though the bones are not completely ossified. METHODS: This is a retrospective review of 77 children/adolescents (mean age 12.7 years) who underwent a lower extremity CT for torsion measurements on a 64-slice scanner. A stepwise dose reduction (70%, 50%, 30% of the original dose) was simulated. Torsion measurements were performed on all image datasets, and image noise, interrater agreement and subjective image quality were evaluated. Effective radiation dose of each original scan was estimated. As proof of concept, 24 children were scanned with an ultra-low-dose protocol, adapted from the 30% dose simulation, and the intra-class correlation coefficient (ICC) was determined. Ethics approval and informed consent were given. RESULTS: Torsion measurements at the simulated 30% dose level had equivalent interrater agreement compared to the 100% dose level (ICC ≥ 0.99 for all locations and dose levels). Image quality of almost all datasets was rated excellent, regardless of dose. The mean sum of the effective dose of the total torsion measurement was reduced by simulation from 0.460/0.490 mSv (boys/girls) at 100% dose to 0.138/0.147 mSv at 30%. The ICC of the proof-of-concept group was as good as that of the simulated 30% dose level. CONCLUSION: Pediatric torsion measurements of the lower extremities can be performed using an ultra-low-dose protocol without compromising diagnostic confidence.
The staging and local management of breast cancer involves the evaluation of the extent and completeness of excision of both the invasive carcinoma component and also the intraductal component or ductal carcinoma in situ. When both invasive ductal carcinoma and coincident ductal carcinoma in situ are present, assessment of the extent and localization of both components is required for optimal therapeutic planning. We have used a mouse model of breast cancer to evaluate the feasibility of applying molecular imaging to assess the local status of cancers in vivo. Multi-tracer positron emission tomography (PET) and magnetic resonance imaging (MRI) characterize the transition from premalignancy to invasive carcinoma. PET tracers for glucose consumption, membrane synthesis, and neoangiogenesis in combination with a Gaussian mixture model-based analysis reveal image-derived thresholds to separate the different stages within the whole-lesion. Autoradiography, histology, and quantitative image analysis of immunohistochemistry further corroborate our in vivo findings. Finally, clinical data further support our conclusions and demonstrate translational potential. In summary, this preclinical model provides a platform for characterizing multistep tumor progression and provides proof of concept that supports the utilization of advanced protocols for PET/MRI in clinical breast cancer imaging.
