Fully Automatic Knee Bone Detection and Segmentation on Three-Dimensional MRI

In the medical sector, three-dimensional (3D) images are commonly used like computed tomography (CT) and magnetic resonance imaging (MRI). The 3D MRI is a non-invasive method of studying the soft-tissue structures in a knee joint for osteoarthritis studies. It can greatly improve the accuracy of segmenting structures such as cartilage, bone marrow lesion, and meniscus by identifying the bone structure first. U-net is a convolutional neural network that was originally designed to segment the biological images with limited training data. The input of the original U-net is a single 2D image and the output is a binary 2D image. In this study, we modified the U-net model to identify the knee bone structures using 3D MRI, which is a sequence of 2D slices. A fully automatic model has been proposed to detect and segment knee bones. The proposed model was trained, tested, and validated using 99 knee MRI cases where each case consists of 160 2D slices for a single knee scan. To evaluate the model’s performance, the similarity, dice coefficient (DICE), and area error metrics were calculated. Separate models were trained using different knee bone components including tibia, femur, patella, as well as a combined model for segmenting all the knee bones. Using the whole MRI sequence (160 slices), the method was able to detect the beginning and ending bone slices first, and then segment the bone structures for all the slices in between. On the testing set, the detection model accomplished 98.79% accuracy and the segmentation model achieved DICE 96.94% and similarity 93.98%. The proposed method outperforms several state-of-the-art methods, i.e., it outperforms U-net by 3.68%, SegNet by 14.45%, and FCN-8 by 2.34%, in terms of DICE score using the same dataset.

NIMG-10. OPTIMIZING IMAGE QUALITY FOR MAGNETIC RESONANCE SIMULATION IN SPINE STEREOTACTIC BODY RADIATION THERAPY

BACKGROUND Currently, the standard MRI sequence for SC imaging in SBRT has been axial 2D T2-weighted Turbo Spin Echo (TSE). Even though 3D T2-weighted sequences such as SPACE (Sampling Perfection with Application optimized Contrasts using different flip angle Evolution) image a whole volume simultaneously and thus offer better reconstruction, they have not been clinically implemented due to their long acquisition times. However, the application of Compressed Sensing (CS) methods on SPACE sequences, achieving clinically acceptable time. METHODS A 3D T2 CS SPACE was obtained and evaluated against the standard 2D TSE for spine SBRT based on a MagPhan RT quality assurance phantom and patients data, analysis was done using the phantom manufacturer software ImageOwl that calculates image distortions by comparing the known position of phantom features to their detected position in the image. RESULTS Results of phantom comparison between 3D T2 and 2D T2 indicate that although the 3D sequence had lower signal-to-noise ratio (SNR) than the 2D sequence, it presented less geometric distortions caused by gradient non-linearities, particularly in the anterior-posterior (A/P) and head-feet (H/F) directions. Distortions caused by chemical shift are in theory smaller for the 3D T2 CS SPACE, amounting to 0.85mm compared to 1.62mm with 2D T2. Between 2D versus 3D MRI defined SC data among 4 patients, average deviation of the centroid point cord contours was 0.08cm. The volume of the cord showed 1cc larger 3D volumes compared to 2D T2. Finally, the mean voxel count overlap coefficient and DICE coefficient was 0.92 and 0.87 respectively. CONCLUSIONS Since 3D MRI is under consideration to replace 2D MRI, it is important to compare SC contours from 3D to 2D MRI and assess their impact on treatment plans. Positive results would pave the path for larger subject cohort evaluation.

Interobserver variability in target volume delineation for CT/MRI simulation and MRI guided adaptive radiotherapy in rectal cancer

Objectives: Quantify target volume delineation uncertainty for CT/MRI simulation and MRI-guided adaptive radiotherapy in rectal cancer. Define optimal imaging sequences for target delineation. Methods: Six experienced radiation oncologists delineated CTVs on CT and 2D and 3D-MRI in three patients with rectal cancer, using consensus contouring guidelines. Tumour GTV (GTVp) was also contoured on MRI acquired week 0 and 3 of radiotherapy. A STAPLE contour was created and volume and interobserver variability metrics were analysed. Results: There were statistically significant differences in volume between observers for CT and 2D-MRI-defined CTVs (p < 0.05). There was no significant difference between observers on 3D-MRI. Significant differences in volume were seen between observers for both 2D and 3D-MRI-defined GTVp at weeks 0 and 3 (p < 0.05). Good interobserver agreement (IOA) was seen for CTVs delineated on all imaging modalities with best IOA on 3D-MRI; median Conformity Index (CI) 0.74 for CT, 0.75 for 2D-MRI and 0.77 for 3D-MRI. IOA of MRI-defined GTVp week 0 was better compared to CT; CI 0.58 for CT, 0.62 for 2D-MRI and 0.7 for 3D-MRI. MRI-defined GTVp IOA week three was worse compared to week 0. Conclusions: Delineation on MRI results in smaller volumes and better IOA week 0 compared to CT. 3D-MRI provides the best IOA in CTV and GTVp. MRI-defined GTVp on images acquired week three showed worse IOA compared to week 0. This highlights the need for consensus guidelines in GTVp delineation on MRI during treatment course in the context of dose escalation MRI-guided rectal boost studies. Advances in knowledge: Optimal MRI sequences for CT/MRI simulation and MRI-guided adaptive radiotherapy in rectal cancer have been defined.

Feasibility of MRI for the evaluation of interosseous ligament vertical segment via subtalar arthroscopy correlation: comparison of 2D and 3D MR images

Background Interosseous ligament vertical segment (IOLV) and calcaneofibular ligament (CFL) have been reported to be important in stabilizing the subtalar joint. Unlike CFL, there is not much information regarding the comparison of MRI results with surgical evaluation of IOLV and the comparison between 2D and 3D MRI on IOLV evaluation. The feasibility of MRI in IOLV evaluation has yet to be reported. The purpose of this study was to evaluate the validity and reliability of MRI in IOLV tear detection via correlation with arthroscopic results. We also compared the diagnostic performance of 2D and 3D MR images. Methods In this retrospective study, 52 patients who underwent subtalar arthroscopy after ankle MRI were enrolled. Arthroscopic results confirmed IOLV tear in 25 cases and intact IOLV in 27 cases. Two radiologists independently evaluated the IOLV tears using only conventional 2D images, followed by isotropic 3D images, and comparison with arthroscopic results. Results Only the 2D sequences interpreted by two readers showed a sensitivity of 64.0–96.0%, a specificity of 29.6–44.4%, a positive predictive value of 51.6–56.4%, and a negative predictive value of 57.1–88.9%. Addition of isotropic 3D sequences changed the sensitivity to 60.0–80.0%, specificity to 63.0–77.8%, positive predictive value to 64.3–76.9%, and negative predictive value to 66.7–80.8%. The overall diagnostic performance of isotropic 3D sequences (AUC values: 0.679–0.816) was higher than that of 2D sequences (AUC values: 0.568–0.647). Inter-observer and intra-observer agreement between the two readers was moderate-to-good for both 2D and 3D sequences. The diagnostic accuracy in 19 patients with tarsal sinus fat obliteration tended to increase from 26.3–42.1% to 57.9–73.7% with isotropic 3D sequences compared with 2D sequences. Conclusions Isotropic 3D MRI was feasible for the assessment of IOLV tear prior to subtalar arthroscopy. Additional 3D sequences showed higher diagnostic accuracy compared with conventional 2D sequences in IOLV evaluation. Isotropic 3D sequences may be more valuable in detecting IOLV tear in case of tarsal sinus fat obliteration.

Prospective Explore of Clinical-Pathological Shrinkage Modes After Neoadjuvant Therapy in Breast Cancer

Purpose: The aim was to explore the clinical-pathological shrinkage modes which oriented by breast conserving surgery (BCS) purpose after neoadjuvant therapy (NAT) using three-dimensional (3D) MRI and pathology reconstruction, in order to guide the individualized selection of BCS candidates and scope of resection after NAT. Methods: From April 2014 to 2018, 104 breast cancer patients underwent operation after NAT were included in this prospective study. All patients underwent MRI examinations before and after NAT. Breast residual tumors were prepared with sub-serial section. The 3D MRI and pathology models were reconstructed with 3D-DOCTOR software. The association and correlation between 3D MRI and pathology models were assessed. The traditional shrinkage modes included the surgical pathology complete response, solitary lesions without surrounding lesions, multinodular lesions, solitary lesions with adjacent spotty lesions and diffuse lesions. Combined with the MD Anderson Cancer Center BCS indications after NAT and traditional shrinkage modes, we derived clinical-pathological shrinkage modes which oriented by BCS purpose: clinical pathological-concentric shrinkage modes (CP-CSM) and clinical pathological-non concentric shrinkage modes (CP-NCSM). The CP-CSM means the longest diameter of residual tumor was less than 50% and ≤2cm in comparison with the primary tumor before NAT. Other shrinkage modes were classified as CP-NCSM. Univariate and multivariate logistic regression analysis was conducted to identify the independent predictive factors of clinical-pathological shrinkage modes. A nomogram was developed based on variables in the final model with p<0.05. Results: Based on the gold standard of 3D pathology reconstruction model-measured shrinkage modes, the accuracy, sensitivity and specificity of 3D MRI reconstruction for predicting the traditional shrinkage modes were 84.6%, 61.9% and 90.4%, respectively (Kappa value=0.497). The accuracy, sensitivity and specificity of 3D MRI reconstruction in predicting clinical-pathological shrinkage modes were 93.3%, 97.0% and 86.5%, respectively (Kappa value=0.850). Multivariate analysis showed that primary tumor stage (OR=2.059, 95%CI: 1.187-3.574), mammographic malignant calcification (OR=3.424, 95%CI: 1.437-8.161), molecular subtypes (OR=0.530, 95%CI: 0.364-0.772) and nodal down-staging after NAT (OR=0.183, 95%CI: 0.067-0.497) were independent predictors of clinical-pathological shrinkage modes (all p<0.05). A nomogram was created based on these four predictors. The AUC value was 0.833 (95％CI: 0.710-0.922). The calibration curve showed a satisfactory fit between the predictive and actual observation. With a median follow-up time of 77 months, the recurrence/metastasis rate in the CP-CSM and CP-NCSM group was 7.1% and 29.4%, respectively (p=0.002). Patients with CP-CSM had a better overall survival and disease-free survival (all p<0.05).Conclusion: The 3D MRI reconstruction after NAT could accurately predict the extent of residual tumor. Combining clinical, imaging, molecular subtypes and NAT efficacy, the nomogram of clinical-pathological shrinkage modes showed sufficient predicting accuracy. And it could help to guide the individualized selection of BCS candidates and scope of resection after NAT, thereby achieve the minimum and effective treatment. However, the applicability of the nomogram still needs to be externally validated. Patients with CP-NCSM after NAT had a worse prognosis.