Essentials
Patient Impact
Introduction
Data Acquisition
Hardware and software considerations
Optimising a DW-MRI protocol
Cancer | N | Treatment | Field strength/Parameter | Gold- standard | Performance | Repeatability/Rep reducibility | QA/QC |
---|---|---|---|---|---|---|---|
Technical validation | |||||||
Locally advanced breast cancer [31] | 4 centres; 54 patients | NA | 1.5T b-values 0, 100, 600, 800 s/mm2
| NA | Evaluation of Gradient nonlinearity correction (GNC) | Mean ΔADC =9.42% without GNC vs 9.41% with GNC (differences up to ±4% in individual pts) | Phantoms: with GNC overall mean error for all sites = 0.6% (without GNC = 9.9%) |
Characterization | |||||||
Malignant musculoskeletal tumours (4 lymphoma, 11 metastases, 26 sarcomas) [32] | 4 centres; 51 patients | NA | 1.5T b-values 0, 1000 s/mm2
Whole tumour ADC | Pathology | Muscle lymphoma showed statistically significant lower ADC values | None | None |
Staging | |||||||
Hodgkin’s or Non- Hodgkin’s Lymphoma [33] | 3 centres; 108 patients | NA | 1.5T Qualitative 2 readers | Bone marrow biopsy, FDG-PET, and follow-up | (Ann Arbor Classification) | =0.51 | None |
Cervical cancer [34] | 2 centres; 68 patients | NA | 3T ADC b-values 0, 150, 500 and 1000 s/mm2
| Pathology | ADC not significantly different in metastatic nodes | 2 Observers No reproducibility reported | Consensus for qualitative evaluation |
Treatment response | |||||||
Solid tumours (phase I) [35] | 2 centres; 13 patients | Combretast atin A4 phosphate and bevacizum ab | 1.5T - ADC total (b-values 0, 50, 100, 250, 500, 750 s/mm2), - ADC high (b- values 100, 250, 500, 750 s/mm2) - ADC low (b-values 0, 50, 100 s/mm2) | Significant increase in median ADC total and ADC high 3 h after the second dose of CA4P | Repeatability (baseline examination) ADC total = 13.3% ADC high = 14.1% ADC low = 62.5% | Sucrose phantom measured at the two sites at 22°C | |
Locally advanced rectal cancer [36] | 3 centres; 38 patients | Preoperativ e chemoradi ation | ?T b-values 0, 1000 s/mm2
| Pathologic al complete response (pCR) rate | ADC increased by 44.5% in pCR group, and decreased by 7.6% in non-pCR group (P = 0.026) | None | None |
Locally Advanced Rectal Cancer [37] | 3 centres; 120 patients | Chemoradi ation | 1.5T - Qualitative (on b-value 1000 s/mm2) -- 3 readers | Pathology tumour regression grade (TRG) | Sens = 52-64% Spec = 89-97% ROC AUC = 0.78- 0.80 | Inter-observer agreement =0.51-0.55 | None |
Locally advanced rectal cancer [38] | 2 centres; 112 patients | Chemoradi ation | 1.5T Qualitative (b-values 1000–1100 s/mm2) DWI post-treatment volume (manual) 2 readers | Pathology tumour regression grade (TRG) | Sens = 70% Spec = 98% ROC AUC = 0.92 | Intraclass corre- lation coefficient (ICC) = 0.72-0.81 | None |
Locally advanced breast cancer [2] | 3 centres; 39 patients | Chemother apy | 3T, b-values 0-800 s/mm2
| PRM ROC AUC at 8-11 days = 0.964 | test-retest for repeatability (13 | thermally controlled diffusion | |
retrospective | Histogram analysis and voxel-based Parametric Response Map (PRM) | Whole tumour ADC ROC AUC at 35 days = 0.825 | pts, 1 centre) ≤ ±0.1x10-3mm2/s. | phantom (1 centre) |
Setting up Quality Assurance: Test Objects
Role of Healthy volunteer studies
Data Storage and Analysis
Data archiving, Transfer and Curation
-
A data storage platform that is resilient, secure and scalable and attached to multiple redundant servers. The object store is a currently popular example [66].
-
A database and associated application program interfaces (APIs) for uploading, querying and downloading data. At present, so-called relational (SQL) databases dominate but the era of Big Data is seeing increasing use made of noSQL concepts.
-
User-facing components that allow a user to access and interact with the data, e.g., a web browser interface and a toolkit of research applications.
Software for image processing
Maintaining quality standards across centres through the life of a trial
QC and Data cleaning
Quality Assurance (QA) | Quality Control (QC) | |
---|---|---|
Why | To prevent errors and defects through planned and systematic actions | To identify and correct defects through a reactive process Benchmarking |
When | Before trial activation | Over duration of trial |
What | • Assure scanner calibration with a test object covering the desired range of ADC • Define minimal quality parameters needed to achieve required accuracy • Assure standardised acquisition by a master guideline • Assure correct acquisition before real patients by a human volunteer scan • Appropriate site training about all requirements and procedures and consider learning curves | • Control of data anonymisation and completeness • Control of data compliance to the imaging guideline - Limited control- randomly selected - Full control- all patients and all time points |
How | • Implement standardized acquisition parameters that take account of variations in image geometry (anatomy, coverage) • Establish trial specific standard operating procedures (SOPs) • Establish trial management plan • Use a secure imaging platform accessible to named personnel at all trial sites | • Check scan quality with pre- defined criteria • Provide feedback to local sites - Retrospectively (by batch or at the end of the trial) - Prospectively (ongoing basis) |
Assessing measurement variability
Good Clinical Practice (GCP)
Reporting considerations for clinical governance
Questions arising from research scan | NIH recommendation (reproduced from Wolf [88]) |
---|---|
Do researchers have an obligation to examine their data for IFs? | ‘It is unrealistic to place on researchers an affirmative duty to search for IFs’ |
What should be done if an IF is detected - should it prompt specialist referral for definitive diagnosis? | ‘Obligation to establish a pathway for handling IFs and communicate that to the Independent ethics committee/review board and research participants’ |
What should the research participant be told? | ‘In many, but not all circumstances, researchers have an obligation to offer to report IFs to participants’ |
What should research protocols and consent forms include relating to IFs, should the right to refuse knowledge of IF be addressed? | ‘Researchers have an obligation to address the possibility of discovering IFs in their protocol and communications with the IRB, also in consent forms and communications with research participants’ |
Key NIH recommendations for addressing IFs: • Plan for the discovery of IFs in study protocol and IRB communication • Plan to verify and evaluate a suspected IF with expert review if necessary • Researchers and IRBs should create and monitor pathways for IFs • Address IFs in the consent process • Plan to determine whether to report IFs, based on likely health importance: a. Strong net benefit to health from reporting IF b. Possible net benefit c. Unlikely net benefit • Address the potential for IFs in future analyses of archived data |
Proposals for future workflow
Factors affecting multicentre DW-MRI variability | Steps to reduce ADC variability |
---|---|
Low SNR of data | Higher field strength, receiver technology (arrays), digital compensation schemes, optimal sequence parameters (including b-values), increased signal averages, interpolation of single pixels/voxels |
Image distortion | Eddy current compensation, improved B0 homogeneity (shimming), increased bandwidth, lower b-values, reduced ETL and matrix |
Ghosting artefacts | Adjust receiver bandwidth and echo-time |
Motion artefacts | Breath-hold, respiratory triggering, cardiac triggering, antiperistaltic agents if necessary |
Statistical errors due to region of interest size | Specify a minimum lesion size for inclusion into the trial; specify ROI size, increase signal averages |
Quality Assurance measures | Standardised test objects, standardised operating procedures for their use and pass/fail criteria |
Test-retest repeatability data | Build test-re-test baseline scans into trial protocol for a subset of patients at each site |
Quality Control measures | Longitudinal review of repeated test object data from each site for the duration of the trial |
Data Transfer, Curation and access | Dedicated server and written standardised procedures within the trial protocol for data anonymisation, transfer to dedicated software platform and access by trial researchers |
Image processing methodology | Robust standardised software (preferably FDA approved or CE marked) that can be accessed by observers from multiple sites to validate reproducibility of results. Standardised segmentation methods (2-D or 3-D, inclusion/exclusion of necrotic areas, manual vs semi- automated or automated ROI definition) |