Table 1. Recommendations for analyzing medical image data in radiology.

These recommendations cover different steps like Design, Analysis and Reporting.

Design: Define research question	-Define research questions -Review related literature and assess the scope of the research -Requirements gathering (required datasets, tools, etc.) -Feasibility assessment and timeline -Refine and finalize research questions and resource requirements
Design: Data and resource curation	-Gather and curate the required resources (data, tools) -Check the quality of imaging and clinical data and perform the appropriate selection
Design: Develop analysis strategy	-Based on available data, define the suitable analysis strategies - Explore and decide the suitable methods and computational approaches: - Feature quantification (engineered vs. deep learning) - Image preprocessing - Data normalization - Supervised vs. unsupervised - Dimensionality reduction - Statistical modeling -Define the analysis flow and timeline -Fix the hypotheses and their evaluation strategies -Fix the training and validation cohorts and ensure no data leakage -Fix the resources to be used
Design: Strive for balance	-Assure that different phenotypic groups are represented appropriately in the training data.
Design: Lock Data	-Lock training and validation cohorts to avoid information leakage. -Ensure that validation data remains locked(unused) until the exploratory analysis and biomarker identification is done on training cohorts.
Analysis: Preprocessing	-Perform the required image pre-processing. -Assess potential batch effects between cohorts and apply correction methods if needed.
Analysis: Radiomic quantification	-Explore and fix the the feature quantification methods. Radiologic characteristics can be quantified, for example, using engineered or deep learned features. -Accordingly fix the feature transformation and data normalization approaches.
Analysis: Biomarker identification	-Explore the feature selection/reduction and machine learning/deep learning modeling approaches using the cross-validation of training cohorts. -Explore different parameter settings and tuning strategies using cross-validation of training data.
Analysis: Lock Methods	-All methods and parameters should be locked (fixed) before applying them to validation data. -A report, testifying that validation data was not seen during the training and exploratory analysis stage, should be sent to the institutional review board (IRB) along with the list of locked methods that will be applied on the validation cohort.
Analysis: Biomarker validation	-Evaluate performance of previously fixed methods in the validation data. -Perform multiple test corrections if applicable -Statistically compare the performance of identified biomarker with conventional clinical markers. -Also, evaluate the complementary additive effect of the identified biomarker on conventional clinical markers and test if there is a significant increase in the performance.
Reporting: Statistics and protocols	-Report all relevant information and parameters related to each statistical test, such as the number of features tested, uncorrected and corrected p-values, effect size, and a rationale for the choice of a model or test. -Report acquisition protocols of the imaging data. Also, report segmentation protocols if these were performed. -List all the software and tools used. -If space is an issue then these details can be provided as a supplementary information to the article or report.
Reporting: Share Data & Methods	-Share the data and methods to the scientific community if feasible. This often brings more reproducibility in science and also increases the overall impact of a publication.