乳癌仍是全球最常見的癌症，其早期診斷與精準治療高度依賴超音波導引介入技術（ultrasound-guided interventions, UGI）。然而，建立穩定的臨床手眼協調與操作能力通常需要累積約20例高風險臨床案例，使傳統「邊做邊學」的住院醫師培訓模式面臨嚴重的訓練瓶頸與患者安全風險。傳統訓練假體，包括大體、動物組織、明膠及矽膠假體，常受限於快速腐敗、過高的針刺摩擦力，或固定且無法調控的物理特性，因而難以重現淺層、深層或小於2公分之等回音病灶等具挑戰性的臨床情境。為改善上述限制，本學位論文開發了一套由聚乙烯醇（polyvinyl alcohol, PVA）水凝膠與有機水凝膠組成之高擬真多層乳房假體系統。本研究利用公開多類別醫學影像資料集中的乳房影像進行三維重建及皮膚輪廓網格擷取，並設計客製化模組化模具，以進行多階段順序澆築及凍融低溫凝膠化。所製備之假體包含皮膚、脂肪及纖維腺體三種解剖層，並嵌入良性囊腫、纖維腺瘤及惡性腫瘤等三種病灶模擬物，以重現不同組織與病灶的超音波影像及觸覺特徵。聲學評估結果顯示，大多數測得之聲速與聲衰減值，以及計算所得之聲阻抗，均落於或接近文獻所報導的人體乳房組織範圍。機械手臂針刺分析進一步評估了假體於剛性、穿刺、切割及摩擦階段的力學反應；與文獻參考曲線相比，其均方根誤差（RMSE）為0.095 N，正規化均方根誤差為5.6%。定量應變彈性造影所得之E-ratio分別為1.37 ± 0.05（良性囊腫）、1.83 ± 0.09（纖維腺瘤）及9.61 ± 0.49（惡性腫瘤），呈現出良性與惡性病灶模擬物之相對剛性差異，並與既有臨床研究所報導的分布趨勢一致。最後，六位臨床專科醫師分別評估兩個乳房假體，共完成12次評估。結果顯示，該假體具有良好的真實感與UGI訓練實用性，四個評估面向之整體前二項評分（top-two-box）滿意度達95%。


Breast cancer remains the most prevalent cancer globally, and its early diagnosis and targeted treatment rely substantially on ultrasound-guided interventions (UGI). However, establishing stable clinical hand–eye coordination and procedural proficiency typically requires approximately 20 high-risk clinical cases, creating a substantial training bottleneck and potential patient-safety concerns within the traditional learning-by-doing residency model. Conventional training phantoms, including cadavers, animal tissues, gelatin, and silicone phantoms, are commonly limited by rapid degradation, excessive needle friction, or fixed and non-tunable physical properties. Consequently, they may not adequately reproduce challenging clinical scenarios involving superficial, deep, or small (<2 cm) isoechoic lesions.To address these limitations, this thesis developed a high-fidelity, multi-layered breast phantom system using polyvinyl alcohol (PVA) hydrogel and organohydrogel formulations. Breast images selected from a publicly available multi-category medical imaging dataset were used for three-dimensional reconstruction and skin-contour mesh extraction. A custom modular mold system was subsequently designed to support multi-stage sequential casting and freeze–thaw cryogelation. The fabricated phantom incorporated three anatomical layers—skin, fat, and fibroglandular tissue—together with three lesion mimics representing a benign cyst, fibroadenoma, and malignant tumor to reproduce tissue- and lesion-specific ultrasound and haptic characteristics.Acoustic evaluation showed that most measured speed-of-sound and attenuation values, together with the calculated acoustic impedance values, fell within or near the ranges reported for human breast tissues in the literature. Robotic needle-insertion analysis further characterized the stiffness, puncture, cutting, and friction responses of the phantom, yielding a root-mean-square error (RMSE) of 0.095 N and a normalized RMSE of 5.6% relative to the literature-derived reference curve. Quantitative strain elastography produced E-ratios of 1.37 ± 0.05 for the benign cyst, 1.83 ± 0.09 for the fibroadenoma, and 9.61 ± 0.49 for the malignant tumor. These results demonstrated relative stiffness differences between the benign and malignant lesion mimics and were consistent with the distributions reported in previous clinical studies. Finally, six clinical specialists each evaluated two breast phantoms, yielding a total of 12 assessments. The results supported the phantom’s perceived realism and UGI training utility, with an overall top-two-box satisfaction rate of 95%, pooled across all items.