With the rapid development of more effective treatments, imaging has become increasingly important not only in diagnosis but also in assisting therapy. For instance, neoadjuvant (pre-operative) chemotherapy (NAC) is one of the major treatment options for locally advanced breast cancer [1]. The assessment of tumor early response to NAC may have a significant impact on patient-specific treatment strategy. Earlier identification of drug resistance can prompt the discontinuation of ineffective therapy (avoiding unnecessary toxicity), and switch to a potentially more effective alternative method (avoiding treatment delays). However, current imaging criteria for evaluating therapeutic response are based on Response Evaluation Criteria in Solid Tumors (RECIST) guidelines [2], which are solely based on tumor shrinkage, a downstream effect of various treatment-induced molecular and cellular changes. Therefore, RECIST criteria are frequently inadequate for assessing early therapeutic response. Currently, there are numerous potentially useful imaging techniques to address this clinical need, including dynamic contrast-enhanced (DCE) [3,4] and diffusion-weighted MRI (DWI) [5,6]. The DWI-derived apparent diffusion coefficient (ADC) has been shown to provide additional and complementary information about tissue cellularity [7] and microstructure that can be used to characterize breast tumors and to monitor their response to treatment [6].

However, clinical trials found ADC only demonstrated moderate prediction performance of treatment response, particularly predicting pathological complete response (pCR) in breast cancer that is associated with long-term outcomes and is a potential surrogate marker for survival [4]. One explanation is that ADC represents averaged diffusion that is influenced by all pathological changes, including those with competing effects [8]. To enhance the specificity of specific pathological changes and improve the predictive performance of treatment response, multiple diffusion MRI (dMRI) based methods have been developed, such as the single-compartmental, time-dependent ADC [8,9] and multi-compartmental biophysical models [[10], [11], [12]] to provide detailed pathophysiological information. These methods, such as IMPULSED [13,14] and VERDICT [15], use both multiple b values and a broad range of diffusion times, and establish multi-compartmental biophysical models to extract microstructural information at the cellular level, such as cell size (effective diameter d), intracellular volume fraction (vin), intra- and extracellular diffusivities (Din and Dex). Previous studies have demonstrated the clinical potential, particularly in breast cancer [16,17] and prostate cancer [18].

However, the current multi-compartment models usually ignore transcytolemmal water exchange (i.e., the water exchange across cell membranes). This is because the entanglement of water exchange and diffusion makes it challenging for biophysical modeling, particularly for finite duration of gradients. However, water exchange causes a mixture of intra- and extracellular water molecules. It is theoretically inappropriate to separate dMRI signals and assume that they are arising from multiple independent compartments. In such cases, neglecting water exchange could result in significant bias in estimating microstructural parameters. For example, it has been found that although cancer cell diameter can still be accurately estimated, the intracellular volume fraction vin is significantly underestimated due to non-negligible water exchange [19]. This prevents these quantitative multi-compartment dMRI methods from providing accurate information on cell density and from estimating cell membrane permeability [20,21].

Meanwhile, there is another type of dMRI method that focuses on transcytolemmal water exchange, including the Kärger model [22,23] and many variant methods such as FEXI [24], NEXI [25], SMEX [26], and diffusion time-dependent kurtosis imaging [27,28]. The Kärger model describes the magnetizations of two Gaussian compartments undergoing water exchange, providing an opportunity to quantify transcytolemmal water exchange rate constants without using contrast agents as in DCE MRI [29,30]. Note that cross-membrane water exchange is associated with tumor malignancy [29] and metabolic activity of cells [31], which has been shown as an important indicator of disease status at the cellular level [[32], [33], [34]].

It is of great interest to perform a multi-parametric MRI to provide a comprehensive characterization of tumors, including cell size, cell density, and transcytolemmal water exchange. Unfortunately, this would significantly increase the total scan time which is not desirable in clinical practice. An ideal approach is to incorporate transcytolemmal water exchange in the multi-compartmental model. This can not only provide comprehensive microstructural information but also enhance the accuracy of estimations of cell size and density [35,36]. Some limited efforts have tried to incorporate the impact of intracellular restricted diffusion into the Kärger model [[37], [38], [39], [40]], such as JOINT [36], enabling simultaneous estimation of cell sizes and water exchange rate constants. However, these Kärger model-based methods are not only unreliable in the case of fast exchange [41] but also usually assume a free (constant diffusivity) or restricted intracellular diffusion, contradictory to the nature of water exchange. More importantly, these methods overestimate the transcytolemmal water exchange rate of large cancer cells due to the edge-enhancement effect [42]. It is plausible to develop a fast and accurate MRI method that provides comprehensive microstructural information including cell size, density, and transcytolemmal water exchange.

In this work, we propose a new dMRI-based microstructural imaging method to address the aforementioned challenges. Our method uses an integrated biophysical model that combines both an extended Kärger model for arbitrary gradient waveforms and the IMPULSED method. Additionally, we introduce a two-mode diffusion model to quantify the influence of transcytolemmal water exchange on intracellular diffusion. To enhance the accuracy of exchange rate constant estimations, we incorporate a practical correction for the edge-enhancement effect. We have named this new imaging method EXCHANGE, as it encompasses water Exchange, Confined (restricted), and Hindered diffusion under Arbitrary Gradient waveform Encodings. Comprehensive validation of EXCHANGE was performed through numerical simulations and retrospective in vitro cell experiments. Furthermore, retrospective in vivo animal experiments and a proof-of-concept patient study demonstrated the clinical potential of the EXCHANGE method. Notably, the total scan time was ∼7 min in monitoring tumor therapeutic response in breast cancer patients with neoadjuvant chemotherapy. The open-sourced sequence and data analysis code will make EXCHANGE readily achievable in clinical trials.