Novel imaging predictors in pediatric lymphoma: radiomics and artificial intelligence. A systematic review

Morphological and molecular imaging are critical for evaluating pediatric lymphoma; image-derived parameters like metabolic tumor volume are highly prognostic. Advanced image analysis methods, such as radiomics and artificial intelligence, can extract relevant parameters and reveal subtle patterns to enhance diagnostic and prognostic evaluations. This systematic review will assess the current evidence of these techniques in PL.

Original, English-language articles published before March 15^th, 2025 and focusing on artificial intelligence or radiomics applications in pediatric lymphoma imaging were reviewed. Papers focused on the analysis of PET, CT, or MRI datasets were considered; from each chosen article, the most representative data, including design, sample size, studies series, and method of analysis, were extracted.

Twelve studies were included; one focused on radiomics, one combined texture analysis with machine learning, and ten explored artificial intelligence applications. Five studies described the use of automatic volume segmentation, demonstrating that artificial intelligence is a reliable and faster alternative to the manual procedure. Four papers evaluated methods for reducing radiation dose, showing that artificial intelligence can reconstruct images of acceptable quality even with a significant decrease in administered radioactivity. Finally, three radiomics/machine learning studies dealt with the differential diagnosis of PL and the stability of the features.

AI and radiomics in PL are still in their early stages but show great promise in automatically extracting important diagnostic parameters, such as the tumor volume, and in delivering sharp diagnostic images with a significant dose reduction to the patient.