Selection of teeth
The present study was approved by the Ethics Committee of the Hospital of Stomatology, Guangzhou Medical University (number KY2017012). Forty human first premolars from an orthodontic tooth extraction in the oral and maxillofacial surgery department were collected extracted, informed written consent was obtained from each patient. Soft and hard tissue residuals on the surfaces of the teeth were removed using an ultrasonic scaler. All teeth had a fully formed apex without any defects or cracks on the surface and had no history of restoration. A curvature of 0-20°, according to Schneider  on buccolingual and mesiodistal radiographs was selected. The selected teeth were of similar dimensions. The evaluation of the sample selection was done by computerized microcomputed tomography, All teeth were numbered and assigned into four groups (n=10/ each group) according to the random number table. CEC groups, Group 1: single-rooted mandibular first premolars with one root canal, Vertucci's classification typeⅠ, Group 2: two-rooted maxillary first premolars with double root canals，Vertucci's classification type Ⅳ. TEC groups,Group 3: single-rooted mandibular first premolars with one root canal, Vertucci's classification typeⅠ, Group 4: two-rooted maxillary first premolars with double root canals, Vertucci's classification type Ⅳ. All the datas were evaluated using CBCT. There was no statistically significant difference (P>0.05) in BL, MD, or tooth root length between the CEC and TEC groups. The teeth were kept in 1% chloramine T trihydrate at room temperature until use.
Manufacture of 3D-printed template
The guided access cavity was prepared using cone-beam computed tomography and optical surface scans. A high-resolution cone-beam computed tomography (CBCT) scan was taken to determine the exact location of the root canal. The drill was virtually superimposed on the root canal to plan the CEC outlines by projecting the access trajectory in each canal orifice that required the least tooth structure removal in Simplant (SIMPLANT, Materialise Dental, Leuven, Belgium) (Fig. 1). The data were then imported into Freeform (Geomagic Freeform, 3D Systems, Morrisville, North Carolina, USA). According to the location of the drill in Simplant, we made a guide template with straight-line pathways into the tooth canal. Besides, we also designed a 3D printed cylindrical lampstand with a 0.2 mm gap to simulate the periodontal ligament for every specimen. The digitally designed template and 3D printed cylindrical lampstand were exported as STL-file and then were sent to a 3D printer (3D System 3510HB, 3D Systems, Morrisville, North Carolina, USA).
Root canal preparation between TECs and CECs
First of all, the teeth were imaged with micro-CT (SkyScan 1172; Bruker micro-CT, Kontich, Belgium) imaging at 20 µm (pretreatment scan) to capture the original canal shape and volume of tooth tissue. In CEC preparation, a 3D-printed template was positioned on the tooth model (Fig. 1), and a guiding sleeve was placed on the hole. CECs were drilled with long diamond burs (MANI SF-11, MANI INC, Japan) at high speed. The CEC access attested the distal and mesial accesses could be directed towards their respective orifices, which kept back the truss of dentin between the cavities. In the TEC group, conventional access cavities were prepared. After initial preparation with pathfile instruments (Dentsply Maillefer, Ballaigues, Switzerland), canals were prepared with 0.04 taper M-Two rotary instruments (VDW company, Munich, Germany) to size 35#. These instruments were used in a standard technique, The canals were irrigated with 3 ml of 5.25% sodium hypochlorite (Guangzhou Hui Fan company, Guangzhou, China) between use of each instrument, and then, each canal was irrigated with 5.25% sodium hypochlorite followed by irrigation for 30 seconds with ultrasonic oscillation tip (K15/21-25, SATELEC, France) coupled with an ultrasound device (SATELEC P5XS, Merignac, France) at power 7. After cleaning and shaping, the teeth were imaged again with micro-CT imaging at 20 µm (posttreatment scan) to capture the instrumented canal shape and volume of tooth tissue for comparative the differences. All canals were obturated with gutta-percha cones (Dentsply Sirona, New York, Pennsylvania, USA) and AH Plus sealer (Dentsply DeTrey, Konstanz, Germany). The thermoplastic continuous wave of condensation technique was used for obturation using a B&L-beta Gutta Percha Heating System (B&L Biotech, Inc, Korea). Smart Dentin Replacement (Dentsply, DE, USA) was used to imitate the lost dentin tissue, and 2 mm composite resin restorative material (SHOFU, Kyoto, Japan) was placed on the canal opening. The teeth were stored in physiological saline at 37°C for one week. After that, each specimen was subjected to micro-CT imaging at 20 µm (finished scan).
Load at fracture
After root canal filling and micro-CT scanning, teeth with a 3D printed cylindrical lampstand were mounted in an Instron Testing machine (E3000, Instron, High Wycombe, UK). The specimens were subjected to 500000 loading cycles in the Instron Testing machine (E3000) axial forces, directed at a 135 angle from the long axis of the tooth , between 5 N-50 N at 15 HZ to simulate approximately 2 years of chewing function [19,20]. After this fatigue phase, the specimens were placed in the Instron Universal Testing machine (E3366, Instron, MA, America). Each tooth was loaded at the central fossa at 135° from the tooth long axis to simulate a maximum bending motion of the tooth at buccal cervical areas . A continuous compressive force was applied with a 2-mm spherical crosshead at 1 mm/min until failure occurred, which was defined as a 25% drop in the applied force  (Fig. 2A). The load at fracture was recorded in Newton (N), and the type of fracture was recorded (Fig. 2B).
After reconstruction with NRecon (Bruker micro CT, Kontich, Belgium) software, the volume of the tooth tissue was analyzed with CT-AN software (Bruker micro CT, Kontich, Belgium), we selected appropriate CT value as the segmentation of tooth volume. After the pretreatment scan and post-treatment scan were aligned in Data Viewer software (Bruker micro CT, Kontich, Belgium), the increased canal volume and surface areas after root canals shaped during the two different access opening procedures were measured using CT-An software. The proportion of untouched canal wall (UCW) in the canals was determined with 3-Matic (Fig. 3), and we measured the sectional section of 1, 3, and 5 mm from the apical and the deviation of the central point in Solid Work (Dassault, France) (Fig. 4). The percentage volume of root filling materials and any voids inside the region of interest were calculated in CT-An software, and all areas without filling within the root canal space were considered voids.
The data were analyzed using IBM SPSS Statistics 16 software (Armonk, NY, USA), it was compared with independent samples T-tests, and the Mann-Whitney U test, P < .05 was considered significant.