Recently developed implantable microdevices can perform multi-drug response assessment of cancer drugs in-vivo, with potential to develop highly optimized personalized cancer treatment strategies. However, minimally invasive/interventional image-guided methods of in-vivo microdevice implantation, securement, and retrieval are needed for broad clinical translation. Here we demonstrate proof-of-concept of an interventional microdevice implantation and retrieval method for personalized drug response assessment, using ex-vivo phantom, ex-vivo tissue, and in-vivo murine models. A method for minimally-invasive microdevice implantation and retrieval was developed, by which a custom-prototyped 6mm retrievable microdevice can be implanted into a live tumor, deliver drugs into 10 discrete regions of adjacent tissue, and retrieved along with the adjacent drug-exposed tissue with a custom-prototyped retrieval needle device to allow in-vivo multi-drug response assessment. Computed tomography (CT) and ultrasound (US)-guided minimally invasive microdevice implantation and retrieval were tested in ex-vivo phantom and tissue models. Successful retrieval was defined as retrieval of the microdevice and adjacent core phantom/tissue sample containing at least 4/10 drug delivery sites. Subsequently, 10 implantation and retrieval trials in phantom models were performed using bi-axial and tri-axial retrieval needles; success rates were calculated and compared using a two-proportion z-test and the number of successfully retrieved drug release sites per microdevice was calculated and compared using a one-tailed independent t-test. Finally, five microdevices, each containing ten reservoirs preloaded with chemotherapy agent Doxorubicin, were implanted into mouse tumors in-vivo, secured for 24-h during drug release, and microdevice/tissue retrieval was performed under ultrasound guidance. Fluorescence microscopy of the retrieved tissue was used to confirm drug delivery and apoptosis staining assessed in-vivo tissue response; correlation of drug release and apoptosis staining were used to assess in-vivo drug efficacy. Image-guided microdevice implantation and retrieval were successful in ex-vivo phantom and tissue models with both US and CT guidance. Bi-axial retrieval success rate was significantly higher than triaxial retrieval in ex-vivo phantom trials (90% vs 50%, z=1.95, P=0.026), and had nonsignificantly higher number of retrieved drug-release sites per microdevice (8.3 vs 7.0, t=1.37, P=0.097). Bi-axial retrieval was successful in all five in-vivo mouse tumor models, and allowed in-vivo drug response assessment at up to ten discrete drug delivery sites per microdevice. An average of 6.8/10 discrete tumor sites containing micro-doses of delivered drug were retrieved per in-vivo attempt (min 5, max 10, std 1.93). Tissue regions of drug delivery, as assessed with fluorescent Doxorubicin drug signal, correlated with regions of apoptosis staining in all in-vivo models, indicating drug efficacy. No bleeding, microdevice migration, or other complications were noted during implantation, 24-h observation, or retrieval. The demonstrated image-guided minimally invasive microdevice implantation and retrieval method is similar to routine outpatient biopsy procedures, obviates the need for surgery, and can be performed at varying depths under CT and/or US guidance. There is potential for this method to enable clinical translation of in-vivo personalized drug response assessment/prediction in a much larger number of patients than currently possible.