OBJECTIVES:Clinical translation of advanced MRI techniques can be hindered by the challenges of performing standardized multicentre imaging trials. This work aims to develop and demonstrate an automated tool for monitoring imaging protocol deviations, enabling corrective action to be taken. METHODS:A Python-based tool, integrated into the imaging repository XNAT, was developed to compare DICOM series with an agreed imaging protocol, highlighting missing series and parameter deviations. This was demonstrated through retrospective analysis of a prospectively acquired dataset from a ten-site whole-body (WB) MRI study of patients with multiple myeloma. The acquired data were compared to the relevant radiological guidelines and to the site-specific imaging protocols agreed for the study. RESULTS:The rate of technical software failure was 0% across 174 examinations from 10 sites. The clinical guidelines were followed in 87.9% of examinations and compliance with the site-specific imaging protocol was greater than 75.0% for all parameters. Common deviations included number of averages for diffusion-weighted imaging (DWI) and repetition time for DWI and Dixon: 85.2%, 81.7%, and 75.1%, respectively. There was a statistically significant correlation between protocol compliance and overall exam radiological image quality. CONCLUSIONS:Repository-integrated software is presented for automated monitoring of imaging protocol compliance to support standardization in multicentre studies and clinical translation. ADVANCES IN KNOWLEDGE:This study presents a novel open-source repository-integrated software tool for automatically monitoring compliance with the expected imaging protocol. Standardized acquisition protocols are crucial in multicentre imaging studies and this tool has the potential to enhance research outcomes and support clinical translation.

Objective. Image quality in whole-body MRI (WB-MRI) may be degraded by faulty radiofrequency (RF) coil elements or mispositioning of the coil arrays. Phantom-based quality control (QC) is used to identify broken RF coil elements but the frequency of these acquisitions is limited by scanner and staff availability. This work aimed to develop a scan-specific QC acquisition and processing pipeline to detect broken RF coil elements, which is sufficiently rapid to be added to the clinical WB-MRI protocol. The purpose of this is to improve the quality of WB-MRI by reducing the number of patient examinations conducted with suboptimal equipment. Approach. A rapid acquisition (14 s additional acquisition time per imaging station) was developed that identifies broken RF coil elements by acquiring images from each individual coil element and using the integral body coil. This acquisition was added to one centre’s clinical WB-MRI protocol for one year (892 examinations) to evaluate the effect of this scan-specific QC. To demonstrate applicability in multi-centre imaging trials, the technique was also implemented on scanners from three manufacturers. Main results . Over the course of the study RF coil elements were flagged as potentially broken on five occasions, with the faults confirmed in four of those cases. The method had a precision of 80% and a recall of 100% for detecting faulty RF coil elements. The coil array positioning measurements were consistent across scanners and have been used to define the expected variation in signal. Significance . The technique demonstrated here can identify faulty RF coil elements and positioning errors and is a practical addition to the clinical WB-MRI protocol. This approach was fully implemented on systems from two manufacturers and partially implemented on a third. It has potential to reduce the number of clinical examinations conducted with suboptimal hardware and improve image quality across multi-centre studies.

Objectives MAchine Learning In MyelomA Response (MALIMAR) is an observational clinical study combining “real-world” and clinical trial data, both retrospective and prospective. Images were acquired on three MRI scanners over a 10-year window at two institutions, leading to a need for extensive curation. Methods Curation involved image aggregation, pseudonymisation, allocation between project phases, data cleaning, upload to an XNAT repository visible from multiple sites, annotation, incorporation of machine learning research outputs and quality assurance using programmatic methods. Results A total of 796 whole-body MR imaging sessions from 462 subjects were curated. A major change in scan protocol part way through the retrospective window meant that approximately 30% of available imaging sessions had properties that differed significantly from the remainder of the data. Issues were found with a vendor-supplied clinical algorithm for “composing” whole-body images from multiple imaging stations. Historic weaknesses in a digital video disk (DVD) research archive (already addressed by the mid-2010s) were highlighted by incomplete datasets, some of which could not be completely recovered. The final dataset contained 736 imaging sessions for 432 subjects. Software was written to clean and harmonise data. Implications for the subsequent machine learning activity are considered. Conclusions MALIMAR exemplifies the vital role that curation plays in machine learning studies that use real-world data. A research repository such as XNAT facilitates day-to-day management, ensures robustness and consistency and enhances the value of the final dataset. The types of process described here will be vital for future large-scale multi-institutional and multi-national imaging projects. Critical relevance statement This article showcases innovative data curation methods using a state-of-the-art image repository platform; such tools will be vital for managing the large multi-institutional datasets required to train and validate generalisable ML algorithms and future foundation models in medical imaging. Key points • Heterogeneous data in the MALIMAR study required the development of novel curation strategies. • Correction of multiple problems affecting the real-world data was successful, but implications for machine learning are still being evaluated. • Modern image repositories have rich application programming interfaces enabling data enrichment and programmatic QA, making them much more than simple “image marts”. Graphical Abstract

BackgroundPneumonitis is a well-described, potentially disabling, or fatal adverse effect associated with both immune checkpoint inhibitors (ICI) and thoracic radiotherapy. Accurate differentiation between checkpoint inhibitor pneumonitis (CIP) radiation pneumonitis (RP), and infective pneumonitis (IP) is crucial for swift, appropriate, and tailored management to achieve optimal patient outcomes. However, correct diagnosis is often challenging, owing to overlapping clinical presentations and radiological patterns.MethodsIn this multi-centre study of 455 patients, we used machine learning with radiomic features extracted from chest CT imaging to develop and validate five models to distinguish CIP and RP from COVID-19, non-COVID-19 infective pneumonitis, and each other. Model performance was compared to that of two radiologists.ResultsModels to distinguish RP from COVID-19, CIP from COVID-19 and CIP from non-COVID-19 IP out-performed radiologists (test set AUCs of 0.92 vs 0.8 and 0.8; 0.68 vs 0.43 and 0.4; 0.71 vs 0.55 and 0.63 respectively). Models to distinguish RP from non-COVID-19 IP and CIP from RP were not superior to radiologists but demonstrated modest performance, with test set AUCs of 0.81 and 0.8 respectively. The CIP vs RP model performed less well on patients with prior exposure to both ICI and radiotherapy (AUC 0.54), though the radiologists also had difficulty distinguishing this test cohort (AUC values 0.6 and 0.6).ConclusionOur results demonstrate the potential utility of such tools as a second or concurrent reader to support oncologists, radiologists, and chest physicians in cases of diagnostic uncertainty. Further research is required for patients with exposure to both ICI and thoracic radiotherapy.

Artificial intelligence (AI) methods applied to healthcare problems have shown enormous potential to alleviate the burden of health services worldwide and to improve the accuracy and reproducibility of predictions. In particular, developments in computer vision are creating a paradigm shift in the analysis of radiological images, where AI tools are already capable of automatically detecting and precisely delineating tumours. However, such tools are generally developed in technical departments that continue to be siloed from where the real benefit would be achieved with their usage. Significant effort still needs to be made to make these advancements available, first in academic clinical research and ultimately in the clinical setting. In this paper, we demonstrate a prototype pipeline based entirely on open-source software and free of cost to bridge this gap, simplifying the integration of tools and models developed within the AI community into the clinical research setting, ensuring an accessible platform with visualisation applications that allow end-users such as radiologists to view and interact with the outcome of these AI tools.

BACKGROUND:The aim of this work is to evaluate the performance of radiomics predictions for a range of molecular, genomic and clinical targets in patients with clear cell renal cell carcinoma (ccRCC) and demonstrate the impact of novel feature selection strategies and sub-segmentations on model interpretability. METHODS:Contrast-enhanced CT scans from the first 101 patients recruited to the TRACERx Renal Cancer study (NCT03226886) were used to derive radiomics classification models to predict 20 molecular, histopathology and clinical target variables. Manual 3D segmentation was used in conjunction with automatic sub-segmentation to generate radiomics features from the core, rim, high and low enhancing sub-regions, and the whole tumour. Comparisons were made between two classification model pipelines: a Conventional pipeline reflecting common radiomics practice, and a Proposed pipeline including two novel feature selection steps designed to improve model interpretability. For both pipelines nested cross-validation was used to estimate prediction performance and tune model hyper-parameters, and permutation testing was used to evaluate the statistical significance of the estimated performance measures. Further model robustness assessments were conducted by evaluating model variability across the cross-validation folds. RESULTS:Classification performance was significant (p < 0.05, H0:AUROC = 0.5) for 11 of 20 targets using either pipeline and for these targets the AUROCs were within ± 0.05 for the two pipelines, except for one target where the Proposed pipeline performance increased by > 0.1. Five of these targets (necrosis on histology, presence of renal vein invasion, overall histological stage, linear evolutionary subtype and loss of 9p21.3 somatic alteration marker) had AUROC > 0.8. Models derived using the Proposed pipeline contained fewer feature groups than the Conventional pipeline, leading to more straightforward model interpretations without loss of performance. Sub-segmentations lead to improved performance and/or improved interpretability when predicting the presence of sarcomatoid differentiation and tumour stage. CONCLUSIONS:Use of the Proposed pipeline, which includes the novel feature selection methods, leads to more interpretable models without compromising prediction performance. TRIAL REGISTRATION:NCT03226886 (TRACERx Renal).

The magnetic field in magnetic resonance imaging-guided radiotherapy (MRgRT) systems influences the three-dimensional (3D) dose deposition and hence the delivered dose distributions. The aim of this study was to investigate the dose-rate dependency and dose response of silicone-based radiochromic dosimeters for photon irradiation in the precense of a magnetic field using a 1.5 T MRgRT system. Additionally, the study aimed to provide a proof of the concept of radiotherapy treatment based on a treatment plan calculated on magnetic resonance imaging rather than a computed tomography (CT) scan. The delivered dose was read out in 3D with an optical CT scanner and the dose distribution was verified using gamma analysis. We found an insignificant dose-rate dependency for dose rates ranging from 3.2 to 5.1 Gy/min and a linear dose response up to 20 Gy. A 3D 3%/3mm gamma analysis showed a pass rate of 95.1%. The dosimeter showed clinical potential for 3D dose verification of MRgRT delivery.

To recognise the many and varied contributions from groups around the world over more than 25 years of research into 3-D radiation dosimetry using optical imaging (primarily optical computed tomography), a systematic review has been conducted with the aim of charting the development of this field. Some 493 publications matched the criteria of the review, and these have been coded with information relating to the category of research, the type of dosimeter material used and the geographical location where the research was conducted. The results of the analysis demonstrate a steady growth in the number of academic groups involved, from 4 at the first DOSGEL meeting in 1999 to over 30 in 2016. The important role played by the DOSGEL and IC3Ddose conference series is demonstrated by the marked biennial trend in publications. The division of publications between categories was 36% materials development, 34% scanner technology, 24% applications, 5% review and 1% teaching. The most common application areas were brachytherapy, proton/particle therapy and IMRT/VMAT. Prior to the introduction of PRESAGE® in 2003, polymer gels were the most popular material, whilst radiochromic dosimeters have dominated in recent years. These and other trends are discussed on the basis of the data presented. A brief discussion of the future of the field, including some personal opinions, ends this brief review.

Monitoring treatment response of soft-tissue sarcomas (STS) following radiotherapy is challenging due to the inherent intratumoral heterogeneity of the disease. Radiomics and deep-learning provide opportunities for the discovery of potent biomarkers of treatment response. Successful response biomarkers must demonstrate good baseline repeatability if they are to be used for personalized treatment. We explore the stability of radiomic features derived from a deep-learning pipeline by determining the pairwise correlation of derived features, and measuring the baseline repeatability of features derived from the Apparent Diffusion-Coefficient maps. We demonstrate that 81/512 features are both independent and stable at repeat baseline measurement.

Background Retroperitoneal sarcomas are tumours with a poor prognosis. Upfront characterisation of the tumour is difficult, and under-grading is common. Radiomics has the potential to non-invasively characterise the so-called radiological phenotype of tumours. We aimed to develop and independently validate a CT-based radiomics classification model for the prediction of histological type and grade in retroperitoneal leiomyosarcoma and liposarcoma.Methods A retrospective discovery cohort was collated at our centre (Royal Marsden Hospital, London, UK) and an independent validation cohort comprising patients recruited in the phase 3 STRASS study of neoadjuvant radiotherapy in retroperitoneal sarcoma. Patients aged older than 18 years with confirmed primary leiomyosarcoma or liposarcoma proceeding to surgical resection with available contrast-enhanced CT scans were included. Using the discovery dataset, a CT-based radiomics workflow was developed, including manual delineation, sub-segmentation, feature extraction, and predictive model building. Separate probabilistic classifiers for the prediction of histological type and low versus intermediate or high grade tumour types were built and tested. Independent validation was then performed. The primary objective of the study was to develop radiomic classification models for the prediction of retroperitoneal leiomyosarcoma and liposarcoma type and histological grade.Findings 170 patients recruited between Oct 30, 2016, and Dec 23, 2020, were eligible in the discovery cohort and 89 patients recruited between Jan 18, 2012, and April 10, 2017, were eligible in the validation cohort. In the discovery cohort, the median age was 63 years (range 27-89), with 83 (49%) female and 87 (51%) male patients. In the validation cohort, median age was 59 years (range 33-77), with 46 (52%) female and 43 (48%) male patients. The highest performing model for the prediction of histological type had an area under the receiver operator curve (AUROC) of 0 center dot 928 on validation, based on a feature set of radiomics and approximate radiomic volume fraction. The highest performing model for the prediction of histological grade had an AUROC of 0 center dot 882 on validation, based on a radiomics feature set.Interpretation Our validated radiomics model can predict the histological type and grade of retroperitoneal sarcomas with excellent performance. This could have important implications for improving diagnosis and risk stratification in retroperitoneal sarcomas.Funding Wellcome Trust, European Organisation for Research and Treatment of Cancer-Soft Tissue and Bone Sarcoma Group, the National Institutes for Health, and the National Institute for Health and Care Research Biomedical Research Centre at The Royal Marsden NHS Foundation Trust and The Institute of Cancer Research.Copyright (c) 2023 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY-NC-ND 4.0 license.

Morten B. Jensen , Lia B. Valdetaro , Peter Balling , Peter S. Skyt , Jørgen B. B. Petersen , Simon J. Doran , Mateusz K. Sitarz and Ludvig P. Muren Danish Centre for Particle Therapy, Aarhus University Hospital, Aarhus, Denmark; Department of Clinical Medicine, Aarhus University, Aarhus, Denmark; Department of Medical Physics, Aarhus University Hospital, Aarhus, Denmark; Department of Physics and Astronomy, Aarhus University, Aarhus, Denmark; Interdisciplinary Nanoscience Center, Aarhus University, Aarhus, Denmark; The Institute of Cancer Research, London, UK

Glioblastomas are the most common and aggressive malignant primary tumor of the central nervous system in adults. The tumours are quite heterogeneous in its shape, texture, and histology. Patients that have been diagnosed with glioblastoma typically have low survival rates and it can take weeks to perform a genetic analysis of an extracted tissue sample. If an effective way to diagnose glioblastomas have been discovered through the use of imaging and AI techniques, this can lead to quality of life improvement for patients through better planning of therapy and surgery required. This work is part of the Brain Tumor Segmentation BraTS 2021 challenge. The challenge is to predict the MGMT promotor methylation status from multi-modal MRI data. We propose a multi-modal late fusion 3D classification network for brain tumor classifcation on 3D MRI images by using all 4 different modalities (T1w, T1wCE, T2w, FLAIR) and also can be extended to include radiomics features or other external features into the network. We also then compare it against 3D classification models trained on each image modality on its own and then ensembled together during inference.

Purpose: XNAT is an informatics software platform to support imaging research, particularly in the context of large, multicentre studies of the type that are essential to validate quantitative imaging biomarkers. XNAT provides import, archiving, processing and secure distribution facilities for image and related study data. Until recently, however, modern data visualisation and annotation tools were lacking on the XNAT platform. We describe the background to, and implementation of, an integration of the Open Health Imaging Foundation (OHIF) Viewer into the XNAT environment. We explain the challenges overcome and discuss future prospects for quantitative imaging studies. Materials and methods: The OHIF Viewer adopts an approach based on the DICOM web protocol. To allow operation in an XNAT environment, a data-routing methodology was developed to overcome the mismatch between the DICOM and XNAT information models and a custom viewer panel created to allow navigation within the viewer between different XNAT projects, subjects and imaging sessions. Modifications to the development environment were made to allow developers to test new code more easily against a live XNAT instance. Major new developments focused on the creation and storage of regions-of-interest (ROIs) and included: ROI creation and editing tools for both contour- and mask-based regions; a “smart CT” paintbrush tool; the integration of NVIDIA’s Artificial Intelligence Assisted Annotation (AIAA); the ability to view surface meshes, fractional segmentation maps and image overlays; and a rapid image reader tool aimed at radiologists. We have incorporated the OHIF microscopy extension and, in parallel, introduced support for microscopy session types within XNAT for the first time. Results: Integration of the OHIF Viewer within XNAT has been highly successful and numerous additional and enhanced tools have been created in a programme started in 2017 that is still ongoing. The software has been downloaded more than 3700 times during the course of the development work reported here, demonstrating the impact of the work. Conclusions: The OHIF open-source, zero-footprint web viewer has been incorporated into the XNAT platform and is now used at many institutions worldwide. Further innovations are envisaged in the near future.

INTRODUCTION:Whole-body MRI (WB-MRI) is recommended by the National Institute of Clinical Excellence as the first-line imaging tool for diagnosis of multiple myeloma. Reporting WB-MRI scans requires expertise to interpret and can be challenging for radiologists who need to meet rapid turn-around requirements. Automated computational tools based on machine learning (ML) could assist the radiologist in terms of sensitivity and reading speed and would facilitate improved accuracy, productivity and cost-effectiveness. The MALIMAR study aims to develop and validate a ML algorithm to increase the diagnostic accuracy and reading speed of radiological interpretation of WB-MRI compared with standard methods.METHODS AND ANALYSIS:This phase II/III imaging trial will perform retrospective analysis of previously obtained clinical radiology MRI scans and scans from healthy volunteers obtained prospectively to implement training and validation of an ML algorithm. The study will comprise three project phases using approximately 633 scans to (1) train the ML algorithm to identify active disease, (2) clinically validate the ML algorithm and (3) determine change in disease status following treatment via a quantification of burden of disease in patients with myeloma. Phase 1 will primarily train the ML algorithm to detect active myeloma against an expert assessment ('reference standard'). Phase 2 will use the ML output in the setting of radiology reader study to assess the difference in sensitivity when using ML-assisted reading or human-alone reading. Phase 3 will assess the agreement between experienced readers (with and without ML) and the reference standard in scoring both overall burden of disease before and after treatment, and response.ETHICS AND DISSEMINATION:MALIMAR has ethical approval from South Central-Oxford C Research Ethics Committee (REC Reference: 17/SC/0630). IRAS Project ID: 233501. CPMS Portfolio adoption (CPMS ID: 36766). Participants gave informed consent to participate in the study before taking part. MALIMAR is funded by National Institute for Healthcare Research Efficacy and Mechanism Evaluation funding (NIHR EME Project ID: 16/68/34). Findings will be made available through peer-reviewed publications and conference dissemination.TRIAL REGISTRATION NUMBER:NCT03574454.

Recurrence occurs in up to 36% of patients treated with curative-intent radiotherapy for NSCLC. Identifying patients at higher risk of recurrence for more intensive surveillance may facilitate the earlier introduction of the next line of treatment. We aimed to use radiotherapy planning CT scans to develop radiomic classification models that predict overall survival (OS), recurrence-free survival (RFS) and recurrence two years post-treatment for risk-stratification. A retrospective multi-centre study of >900 patients receiving curative-intent radiotherapy for stage I-III NSCLC was undertaken. Models using radiomic and/or clinical features were developed, compared with 10-fold cross-validation and an external test set, and benchmarked against TNM-stage. Respective validation and test set AUCs (with 95% confidence intervals) for the radiomic-only models were: (1) OS: 0.712 (0.592–0.832) and 0.685 (0.585–0.784), (2) RFS: 0.825 (0.733–0.916) and 0.750 (0.665–0.835), (3) Recurrence: 0.678 (0.554–0.801) and 0.673 (0.577–0.77). For the combined models: (1) OS: 0.702 (0.583–0.822) and 0.683 (0.586–0.78), (2) RFS: 0.805 (0.707–0.903) and 0·755 (0.672–0.838), (3) Recurrence: 0·637 (0.51–0.·765) and 0·738 (0.649–0.826). Kaplan-Meier analyses demonstrate OS and RFS difference of >300 and >400 days respectively between low and high-risk groups. We have developed validated and externally tested radiomic-based prediction models. Such models could be integrated into the routine radiotherapy workflow, thus informing a personalised surveillance strategy at the point of treatment. Our work lays the foundations for future prospective clinical trials for quantitative personalised risk-stratification for surveillance following curative-intent radiotherapy for NSCLC.

Introduction The spleen is a lymphoid organ and we hypothesize that clinical benefit to immunotherapy may present with an increase in splenic volume during treatment. The purpose of this study was to investigate whether changes in splenic volume could be observed in those showing clinical benefit versus those not showing clinical benefit to pembrolizumab treatment in non-small cell lung cancer (NSCLC) patients. Materials and methods In this study, 70 patients with locally advanced or metastatic NSCLC treated with pembrolizumab; and who underwent baseline CT scan within 2 weeks before treatment and follow-up CT within 3 months after commencing immunotherapy were retrospectively evaluated. The splenic volume on each CT was segmented manually by outlining the splenic contour on every image and the total volume summated. We compared the splenic volume in those achieving a clinical benefit and those not achieving clinical benefit, using non-parametric Wilcoxon signed-rank test. Clinical benefit was defined as stable disease or partial response lasting for greater than 24 weeks. A p-value of <0.05 was considered statistically significant. Results There were 23 responders and 47 non-responders based on iRECIST criteria and 35 patients with clinical benefit and 35 without clinical benefit. There was no significant difference in the median pre-treatment volume (175 vs 187 cm 3 , p = 0.34), post-treatment volume (168 vs 167 cm 3 , p = 0.39) or change in splenic volume (-0.002 vs 0.0002 cm 3 , p = 0.97) between the two groups. No significant differences were also found between the splenic volume of patients with partial response, stable disease or progressive disease (p>0.017). Moreover, there was no statistically significant difference between progression-free survival and time to disease progression when the splenic volume was categorized as smaller or larger than the median pre-treatment or post-treatment volume (p>0.05). Conclusion No significant differences were observed in the splenic volume of those showing clinical benefit versus those without clinical benefit to pembrolizumab treatment in NSCLC patients. CT splenic volume cannot be used as a potentially simple biomarker of response to immunotherapy.

Radiomics refers to the extraction of mineable data from medical imaging and has been applied within oncology to improve diagnosis, prognostication, and clinical decision support, with the goal of delivering precision medicine. The authors provide a practical approach for successfully implementing a radiomic workflow from planning and conceptualization through manuscript writing. Applications in oncology typically are either classification tasks that involve computing the probability of a sample belonging to a category, such as benign versus malignant, or prediction of clinical events with a time-to-event analysis, such as overall survival. The radiomic workflow is multidisciplinary, involving radiologists and data and imaging scientists, and follows a stepwise process involving tumor segmentation, image preprocessing, feature extraction, model development, and validation. Images are curated and processed before segmentation, which can be performed on tumors, tumor subregions, or peritumoral zones. Extracted features typically describe the distribution of signal intensities and spatial relationship of pixels within a region of interest. To improve model performance and reduce overfitting, redundant and nonreproducible features are removed. Validation is essential to estimate model performance in new data and can be performed iteratively on samples of the dataset (cross-validation) or on a separate hold-out dataset by using internal or external data. A variety of noncommercial and commercial radiomic software applications can be used. Guidelines and artificial intelligence checklists are useful when planning and writing up radiomic studies. Although interest in the field continues to grow, radiologists should be familiar with potential pitfalls to ensure that meaningful conclusions can be drawn. Online supplemental material is available for this article. Published under a CC BY 4.0 license.

Whole body magnetic resonance imaging (WB-MRI) is the recommended modality for diagnosis of multiple myeloma (MM). WBMRI is used to detect sites of disease across the entire skeletal system, but it requires significant expertise and is time-consuming to report due to the great number of images. To aid radiological reading, we propose an auxiliary task-based multiple instance learning approach (ATMIL) for MM classification with the ability to localize sites of disease. This approach is appealing as it only requires patient-level annotations where an attention mechanism is used to identify local regions with active disease. We borrow ideas from multi-task learning and define an auxiliary task with adaptive reweighting to support and improve learning efficiency in the presence of data scarcity. We validate our approach on both synthetic and real multi-center clinical data. We show that the MIL attention module provides a mechanism to localize bone regions while the adaptive reweighting of the auxiliary task considerably improves the performance.

Genetic intra-tumour heterogeneity fuels clonal evolution, but our understanding of clinically relevant clonal dynamics remain limited. We investigated spatial and temporal features of clonal diversification in clear cell renal cell carcinoma through a combination of modelling and real tumour analysis. We observe that the mode of tumour growth, surface or volume, impacts the extent of subclonal diversification, enabling interpretation of clonal diversity in patient tumours. Specific patterns of proliferation and necrosis explain clonal expansion and emergence of parallel evolution and microdiversity in tumours. In silico time-course studies reveal the appearance of budding structures before detectable subclonal diversification. Intriguingly, we observe radiological evidence of budding structures in early-stage clear cell renal cell carcinoma, indicating that future clonal evolution may be predictable from imaging. Our findings offer a window into the temporal and spatial features of clinically relevant clonal evolution.