MRI radiomics predicts photoreceptorness status in retinoblastoma

An MRI radiomics model predicted loss of photoreceptorness (a tumor differentiation marker) in retinoblastoma with an AUC of 0.81 in an independent testing cohort, providing a potential non-invasive biomarker for a cancer where biopsy is contraindicated.

• The best model, a random forest with recursive feature elimination, achieved an AUC of 0.83 during cross-validation in the 29-patient training cohort.
• Photoreceptorness reflects photoreceptor-related gene expression; this study shows pretreatment T2-weighted MRI features may correlate with this molecular characteristic.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.