Radiological and Biological Dictionary of Radiomics Features: Addressing Understandable AI Issues in Personalized Prostate Cancer; Dictionary Version PM1.0

Artificial intelligence (AI) has the potential to enhance medical diagnostics, but its clinical application is often limited by issues related to interpretability. This study bridges the gap by associating standardized quantitative imaging features, known as radiomics features ($\mathbf{R F}$), derived from medical images with established clinical frameworks such as PI-RADS, ensuring that AI models are both interpretable and aligned with clinical practice. A team comprising two medical physicists, a physician, a radiologist, and two MDs collaboratively created a radiological/biological dictionary that connects the visual semantic features of PI-RADS with radiomics features, promoting a unified understanding among medical and AI professionals. In this study, six interpretable and seven complex classifiers, paired with nine feature selection algorithms focused on risk factors, were applied to segmented lesions in multiparametric prostate MRI sequences (T2-weighted (T2WI), diffusion-weighted (DWI), and apparent diffusion coefficient (ADC) imaging) to predict UCLA scores. The dictionary was then used to interpret the most predictive models. By combining T2WI, DWI, and ADC with FSAs such as ANOVA F-test, Correlation Coefficient, and Fisher Score, and using logistic regression, key features were identified: the 90th percentile from T2WI, indicating hypointensity linked to prostate cancer risk; variance from T2WI, representing lesion heterogeneity; shape metrics like Least Axis Length and Surface Area to Volume ratio from ADC, describing lesion shape and compactness; and Run Entropy from ADC, which reflects texture consistency. This approach achieved an average accuracy of$0.78 \pm 0.01$, significantly surpassing single-sequence methods ($\mathbf{p}$-value$<0.05$). The developed Prostate-MRI dictionary (PM1.0) provides a shared framework, fostering collaboration between clinicians and AI developers to create trustworthy, interpretable AI solutions that support reliable clinical decision-making.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.