How to avoid describing your radiological research study incorrectly

How to Design a Study and Write a Grant in Radiology?

Radiology as a specialty has grown immensely over the last few decades. The advancements in diagnostic radiology are facilitated by state-of-the-art equipment and a revolution in computation. Parallelly, interventional radiology has benefited and interventional radiologists are at the forefront of several therapeutic procedures. One of the key factors in the growth of radiology is high-quality research. Although the formulation of research questions and hypotheses is similar, the research methodology in diagnostic radiology differs from other specialties. High-quality research requires funds. Thus, it is essential for academic radiologists pursuing research in radiology to be able to write successful grants and secure funds. In this review article, we discuss the strategies to design a study and write successful grants.

One model to unite them all: Personalized federated learning of multi-contrast MRI synthesis

Curation of large, diverse MRI datasets via multi-institutional collaborations can help improve learning of generalizable synthesis models that reliably translate source- onto target-contrast images. To facilitate collaborations, federated learning (FL) adopts decentralized model training while mitigating privacy concerns by avoiding sharing of imaging data. However, conventional FL methods can be impaired by the inherent heterogeneity in the data distribution, with domain shifts evident within and across imaging sites. Here we introduce the first personalized FL method for MRI Synthesis (pFLSynth) that improves reliability against data heterogeneity via model specialization to individual sites and synthesis tasks (i.e., source-target contrasts). To do this, pFLSynth leverages an adversarial model equipped with novel personalization blocks that control the statistics of generated feature maps across the spatial/channel dimensions, given latent variables specific to sites and tasks. To further promote communication efficiency and site specialization, partial network aggregation is employed over later generator stages while earlier generator stages and the discriminator are trained locally. As such, pFLSynth enables multi-task training of multi-site synthesis models with high generalization performance across sites and tasks. Comprehensive experiments demonstrate the superior performance and reliability of pFLSynth in MRI synthesis against prior federated methods.

Analysis of mammograms using artificial intelligence to predict response to neoadjuvant chemotherapy in breast cancer patients: proof of concept

Objectives

In this proof of concept study, a deep learning–based method for automatic analysis of digital mammograms (DM) as a tool to aid in assessment of neoadjuvant chemotherapy (NACT) treatment response in breast cancer (BC) was examined.

Methods

Baseline DM from 453 patients receiving NACT between 2005 and 2019 were included in the study cohort. A deep learning system, using the aforementioned baseline DM, was developed to predict pathological complete response (pCR) in the surgical specimen after completion of NACT. Two image patches, one extracted around the detected tumour and the other from the corresponding position in the reference image, were fed into a classification network. For training and validation, 1485 images obtained from 400 patients were used, and the model was ultimately applied to a test set consisting of 53 patients.

Results

A total of 95 patients (21%) achieved pCR. The median patient age was 52.5 years (interquartile range 43.7–62.1), and 255 (56%) were premenopausal. The artificial intelligence (AI) model predicted the pCR as represented by the area under the curve of 0.71 (95% confidence interval 0.53–0.90; p = 0.035). The sensitivity was 46% at a fixed specificity of 90%.

Conclusions

Our study describes an AI platform using baseline DM to predict BC patients’ responses to NACT. The initial AI performance indicated the potential to aid in clinical decision-making. In order to continue exploring the clinical utility of AI in predicting responses to NACT for BC, further research, including refining the methodology and a larger sample size, is warranted.

Key Points

• We aimed to answer the following question: Prior to initiation of neoadjuvant chemotherapy, can artificial intelligence (AI) applied to digital mammograms (DM) predict breast tumour response?

• DMs contain information that AI can make use of for predicting pathological complete (pCR) response after neoadjuvant chemotherapy for breast cancer.

• By developing an AI system designed to focus on relevant parts of the DM, fully automatic pCR prediction can be done well enough to potentially aid in clinical decision-making.