146 Chapter 8 methods of AI modelling. Limited research is available dedicated to the role of more basic visual image interpretation. Van Griethuysen et al. were one of the first to develop a scoring system to predict response primarily based on visual staging and risk assessment by radiologists on baseline MRI [32]. In Chapter 7 we evaluated this scoring system in a multicentre and multireader study setting and compared it with two simplified adaptations of the same method. We reached a performance of AUC 0.71-0.74 to predict if patients are likely to achieve a (near-)complete response, with most favourable results in terms of interreader agreement and preference for a relatively simple 4-point risk score based on a combination of high risk T-stage, mesorectal fascia (MRF) invasion, extramural vascular invasion (EMVI), and nodal involvement. These results are encouraging, but not sufficient as a basis for clinical decision making. Response to chemoradation is a complex process that is influenced by a variety of factors which we cannot hope to capture by imaging alone. There have been some reports that have shown promising results for combining image based prediction models with other clinical, histopathological, immunohistochemical and genetic predictors of response [33–39]. Moving forward, it will be crucial to incorporate these additional factors into predictive models to enhance our ability to accurately predict response to anti-cancer treatment. By continuing to build on the existing research and incorporating new data and technologies, we can work towards developing more accurate prediction models to further boost personalized therapy in rectal cancer. Conclusions and future outlook With this thesis we have clearly demonstrated the value of a practical web-based tool to enable large scale testing and validation of visual diagnostic methods and staging templates, while at the same time offering new standards for online feedback, training and teaching. We have shown that despite efforts to standardize the radiological staging of rectal cancer there are still certain pitfalls that should be accounted for in future guideline and template updates and for which dedicated (online) teaching can be of added value. In the era of organ-preservation, radiological restaging after neoadjuvant treatment should be tailored towards those clinical questions that impact patient selection and should incorporate DWI as an integral part of the response assessment. Finally, we showed that radiologists’ experience as well as image quality significantly affect diagnostic performance, emphasizing the need for state-of-the-art imaging and dedicated radiologist training. While iScore was designed and tested within the clinical context of rectal cancer, its use can obviously be expanded to promote research and education in other tumour types and disease entities. Future research should also focus on exploring opportunities to connect iScore with other data sharing platforms, and to incorporate links to AI tools to promote further clinical validation and implementation of such tools.
RkJQdWJsaXNoZXIy MjY0ODMw