MRI phased-array coil can achieve large FOV and high SNR simultaneously if coils in an array can be sufficiently decoupled. Coil decoupling can be achieved by either overlapping coil elements¹ or adding inductors and capacitors between coils². Note that the anatomical structure can vary significantly among clinical applications, a flexible coil array allowing for a higher filling factor for different subjects is desired^3-4. However, decoupling by overlapping coil restricts the mechanical flexibility of the coil array. Coil decoupling by adding inductors and capacitors has the disadvantage of ohmic contact loss. Here we propose to decouple between coils in an phased-array by using circumferential shielding stacked on top of the coil (Fig.1). We hypothesize that such a stacked structure is particularly useful when the coil array may be mechanically bended, because clear gaps between coils allows bending. Also the stacked structure permits exchanged for extendable anatomical range. Here we studied two types of circumferential shielding (Fig.2) and compared their performance using full-wave electromagnetic simulation. Simulation inaccuracies and variations of components can be easily compensated by further fine-tuning the practical coils in the future. Our results demonstrate that both types achieved good isolation between coils (S₂₁ lower than -17.1dB) and provided localized sensitivity B₁ regardless of the array was flat or bended.The simulation setup was illustrated in Fig.2⁵. Two decoupled PEC loops were located above the center of the RF coils at right in order to evaluate isolation between coils using S₁₂. A liquid-filled cylindrical phantom was included to simulate the loading, which reduced the quality factor (Q) such speeded up to meet convergence criteria. We simulated two side-by-side circular coils (diameter = 50 mm, width = 5 mm, thickness = 0.02mm) built by copper sheet adhering on FR4 substrate with a 2 mm gap between coils. Each circular coil used three evenly distributed tuning capacitors. The capacitor value was tuned to drive resonance frequency to approach the ¹H Larmor frequency (125 MHz) at 3T. The electromagnetic simulation was done using HFSS (Ansys, Canonsburg, USA). All analysis parameters used the following parameters: max ΔS = 0.02, minimal coverage pass = 2, mixed-order basis functions, sweep-type = fast. Two types of circumferential shielding were modeled: type-I (Fig.2C) was a single strip with width = 8 mm⁶; Type-II (Fig.2D) consisted of two parallel rings (width = 1mm, gap = 8mm). The sizes of both shielding were adjusted to achieve the best decoupling. To demonstrate the flexibility as changing the relative position of two coils on decoupling, each coil array was either placed on the flat top of the phantom or bended to fit to the curvature of the phantom. Maps of |B₁| were evaluated by formulas described in Ref.7.Fig.3 shows |B₁| maps when coils were placed on the flat top of the phantom without and with decoupling shielding, respectively. Coupled B₁ field was apparent beneath the RF coils at left. The coupled |B₁| reduced to 3.1% and 6.8% with type-I and type-II shielding, respectively. The max |B₁| beneath the RF coil at right and S₂₁ with type-I and type-II shielding were {6.31x10^-6(a.u.), -20.4(dB)} and {8.41x10^-6(a.u.), -19.6(dB)}, respectively. Fig.4 shows |B₁| maps when coils were bended to fit to the curvature of a 200mm-diameter cylinder. The coupled B₁ field without shielding was about 85% of flat-top results in Fig.3. With type-I and type-II shielding, the coupled |B₁| reduced to 2.5% and 4.9% and the resonant frequency drifted ≤ 0.75 MHz. The max |B₁| beneath the RF coil at right and S₂₁ were {3.74x10^-6(a.u.), -18.2(dB)} and {5.66x10^-6(a.u.), -17.2(dB)}, respectively. Figs.5 shows the |B₁| maps on the curvatures of 180mm and 220mm-diameter cylinders. Since bending an array inevitably changed the relative positions of two coils, the mutual inductance was no longer the same and the coupling between coils was significantly different after such a geometrical deformation. However, both types of circumferential shielding had similar B₁ distribution and penetration-depth, successfully maintained the image quality in these scenarios.We used numerical simulation to demonstrate that a flexible coil array consisting of circumferential shielding stacked on circular RF coils can be used to acquire high quality MRI even when the array was mechanically bended. While we only demonstrated an array of two coils, we expect that more channels with circumferential shielding can be integrated into an array in order to increase FOV while maintaining the same decoupling between coils. Also our simulation results suggest that type-II (two-parallel-ring) structure performs better than type-I (single-strip). Future work will empirically construct these coils and validate their performance In Vivo.