Personalised estimation of the arterial input function for improved pharmacokinetic modelling of colorectal cancer using dceMRI

dceMRI is becoming a key modality for tumour characterisation and monitoring of response to therapy, because of the ability to identify the underlying tumour physiology. Pharmacokinetic PK models relate the contrast enhancement seen in dceMRI to physiological parameters but require accurate measurement of the AIF, the time-dependant contrast concentration in blood plasma. In this study, a novel method is introduced that overcomes the challenges of direct AIF measurement, by automatically estimating the AIF from the tumour tissue. This approach was evaluated on synthetic data 10% noise and achieved a relative error in K trans and k ep of 11.8 ±3.5% and 25.7 ±4.7 %, respectively, compared to 41 ±15 % and 60 ±32 % using a population model. The method improved the fit of the PK model to clinical colorectal cancer cases, was stable for independent regions in the tumour, and showed improved localisation of the PK parameters. This demonstrates that personalised AIF estimation can lead to more accurate PK modelling.