AFFILIAZIONE
medipass spa
AUTORE PRINCIPALE
Dr. De Summa Marco
VALUTA IL CHALLENGE
Registrazione obbligatoria. Una valutazione consentita
GRUPPO DI LAVORO
Ing. Spinosa Sandro medipass spa
AREA TEMATICA
Applicazioni di intelligenza artificiale in sanità
ABSTRACT
Review of studies on denoising PET-CT images reconstructed with the help of artificial intelligence. Assessment of the feasibility of AI instruments in the clinical routine. The study reviewed the impact of AI-based denoising on PET radiomics, focusing on its effects compared to EARL1 gaussian filtering in a clinical setting with 113 patients. The research involved segmenting 101 FDG-capturing lesions and extracting textural features from PET images using pyradiomics. The comparison between the original PET images and those processed by AI and EARL1 denoising was assessed using the concordance correlation coefficient. A high ratio of concordant features (86.5%) with AI denoising, significantly more than EARL1 denoising (44.2%). Consistency in SUVpeak, SUVmean, SUVmedian values between both denoising methods, contrasting with the stability of SUVmax and SUVmin in AI denoising only.- AI filtering predominantly influenced lower intensity regions, whereas EARL1 gaussian filters affected both low and high intensity regions similarly. In lesions, a significant difference (79%) in texture features was noted between AI denoised and original PET images.The study highlighted that AI-based denoising retains more lesion texture information compared to Gaussian filtering. It suggests that AI denoising can adapt to different tissue types, preserving essential information, making it a promising approach in PET imaging. The overall conclusions pave the way for implementing AI-based tools in routine clinical practice by reconstructing images with low scan times or reduced radiopharmaceutical dose; these images can be used in radiomics studies. Artificial intelligence in PET-CT is a very promising approach as it adapts the denoising for noise versus non-noise
components preserving information where it should. The impact on healthcare of patient is promising, especially in paediatric population. Thus, this approach may reduce the dose and consequently the cost of radiopharmaceuticals and increases the productivity and efficiency.
components preserving information where it should. The impact on healthcare of patient is promising, especially in paediatric population. Thus, this approach may reduce the dose and consequently the cost of radiopharmaceuticals and increases the productivity and efficiency.