Multilingual RECIST classification of radiology reports using supervised learning.

Objectives: The objective of this study is the exploration of Artificial Intelligence and Natural Language Processing techniques to support the automatic assignment of the four Response Evaluation Criteria in Solid Tumors (RECIST) scales based on radiology reports. We also aim at evaluating how languages and institutional specificities of Swiss teaching hospitals are likely to affect the quality of the classification in French and German languages. Methods: In our approach, 7 machine learning methods were evaluated to establish a strong baseline. Then, robust models were built, fine-tuned according to the language (French and German), and compared with the expert annotation. Results: The best strategies yield average F1-scores of 90% and 86% respectively for the 2-classes (Progressive/Non-progressive) and the 4-classes (Progressive Disease, Stable Disease, Partial Response, Complete Response) RECIST classification tasks. Conclusions: These results are competitive with the manual labeling as measured by Matthew's correlation coefficient and Cohen's Kappa (79% and 76%). On this basis, we confirm the capacity of specific models to generalize on new unseen data and we assess the impact of using Pre-trained Language Models (PLMs) on the accuracy of the classifiers.

Increased Construct Stiffness With Meniscal Repair Sutures and Devices Increases the Risk of Cheese-Wiring During Biomechanical Load-to-Failure Testing.

BACKGROUND: Cheese-wiring, the suture that cuts through the meniscus, is a well-known issue in meniscal repair. So far, contributing factors are neither fully understood nor sufficiently studied. HYPOTHESIS/PURPOSE: To investigate whether the construct stiffness of repair sutures and devices correlates with suture cut-through (cheese-wiring) during load-to-failure testing. STUDY DESIGN: Controlled laboratory study. METHODS: In 131 porcine menisci, longitudinal bucket-handle tears were repaired using either inside-out sutures (n = 66; No. 0 Ultrabraid, 2-0 Orthocord, 2-0 FiberWire, and 2-0 Ethibond) or all-inside devices (n = 65; FastFix360, Omnispan, and Meniscal Cinch). After cyclic loading, load-to-failure testing was performed. The mode of failure and construct stiffness were recorded. A receiver operating characteristic curve analysis was performed to define the optimal stiffness threshold for predicting meniscal repair failure by cheese-wiring. The 2-tailed t test and analysis of variance were used to test significance. RESULTS: Loss of suture fixation was the most common mode of failure in all specimens (58%), except for the Omnispan, which failed most commonly because of anchor pull-through. The Omnispan demonstrated the highest construct stiffness (30.8 ± 3.5 N/mm), whereas the Meniscal Cinch (18.0 ± 8.8 N/mm) and Ethibond (19.4 ± 7.8 N/mm) demonstrated the lowest construct stiffness. The Omnispan showed significantly higher stiffness compared with the Meniscal Cinch (P < .001) and Ethibond (P = .02), whereas the stiffness of the Meniscal Cinch was significantly lower compared with that of the FiberWire (P = .01), Ultrabraid (P = .04), and FastFix360 (P = .03). While meniscal repair with a high construct stiffness more often failed by cheese-wiring, meniscal repair with a lower stiffness failed by loss of suture fixation, knot slippage, or anchor pull-through. Meniscal repair with a stiffness >26.5 N/mm had a 3.6 times higher risk of failure due to cheese-wiring during load-to-failure testing (95% CI, 1.4-8.2; P < .0001). CONCLUSION: Meniscal repair using inside-out sutures and all-inside devices with a higher construct stiffness (>26.5 N/mm) was more likely to fail through suture cut-through (cheese-wiring) than that with a lower stiffness (≤26.5 N/mm). CLINICAL RELEVANCE: This is the first study investigating the impact of construct stiffness on meniscal repair failure by suture cut-through (cheese-wiring).

All-inside meniscal repair devices compared with their matched inside-out vertical mattress suture repair: introducing 10,000 and 100,000 loading cycles.

BACKGROUND: All-inside arthroscopic meniscal repairs are favored by most clinicians because of their lower complication rate and decreased morbidity compared with inside-out techniques. Until now, only 1000 cycles have been used for biomechanical testing. HYPOTHESIS: All-inside meniscal repairs will show inferior biomechanical response to cyclic loading (up to 100,000 cycles) and load-to-failure testing compared with inside-out suture controls. STUDY DESIGN: Controlled laboratory study. METHODS: Bucket-handle tears in 72 porcine menisci were repaired using the Omnispan and Fast-Fix 360 (all-inside devices) and Orthocord 2-0 and Ultrabraid 2-0 sutures (matched controls). Initial displacement, displacement after cyclic loading (100, 500, 1000, 2000, 5000, 10,000, and 100,000 cycles) between 5 and 20 N, ultimate load to failure, and mode of failure were recorded, as well as stiffness. RESULTS: Initial displacement and displacement after cyclic loading were not different between the groups. The Omnispan repair demonstrated the highest load-to-failure force (mean ± SD, 151.3 ± 21.5 N) and was significantly stronger than all the other constructs (Orthocord 2-0, 105.5 ± 20.4 N; Ultrabraid 2-0, 93.4 ± 22.5 N; Fast-Fix 360, 76.6 ± 14.2 N) (P < .0001 for all). The Orthocord vertical inside-out mattress repair was significantly stronger than the Fast-Fix 360 repair (P = .003). The Omnispan (30.8 ± 3.5 N/mm) showed significantly higher stiffness compared with the Ultrabraid 2-0 (22.9 ± 6.9 N/mm, P < .0001) and Fast-Fix 360 (23.7 ± 3.9 N/mm, P = .001). The predominant mode of failure was suture failure. CONCLUSION: All-inside meniscal devices show comparable biomechanical properties compared with inside-out suture repair in cyclic loading, even after 100,000 cycles. CLINICAL RELEVANCE: Eight to 10 weeks of rehabilitation might not pose a problem for all repairs in this worst-case scenario.