• artificial intelligence
• chest computed tomography imaging
• image quality evaluation
• phantom study
• radiomic feature robustness
• radiomics
• ultra-low-dose CT protocols

Lung cancer is the leading cause of cancer death worldwide and its diagnosis is complex; in fact, lung cancer is often asymptomatic and its first radiological sign is the presence of sometimes very small lung nodules, detectable only through Computed Tomography (CT). A lung cancer screening program would be useful in reducing the mortality rate but many problems delay its implementation, among them the periodic exposure of potentially healthy patients to ionizing radiation and the difficulties related to nodule detection. One prospect for improving the ability to detect lung nodules is provided by combining Artificial Intelligence-based Computer-Aided Detection systems with Radiomics, but the dependence of radiomic features on image acquisition parameters limits the feasibility of multi-centric studies and thus the generalizability of radiomic models.
In this master thesis, the problems related to the radiation dose reduction in chest CT scans and the radiomic features robustness are addressed using clinical CT scanners, clinical protocols and two different phantoms: the commercial Catphan phantom and the in-house developed Radiomik phantom. The results obtained indicate that, with the use of high iterative reconstruction blending levels, it could be possible to design ultra low dose CT acquisition protocols without degrading the diagnostic information, and that an image quality-based harmonization strategy could improve radiomic features robustness in multi-centric studies.