Assessing robustness and generalization of a deep neural network for brain MS lesion segmentation on real-world data.

OBJECTIVES: Evaluate the performance of a deep learning (DL)-based model for multiple sclerosis (MS) lesion segmentation and compare it to other DL and non-DL algorithms. METHODS: This ambispective, multicenter study assessed the performance of a DL-based model for MS lesion segmentation and compared it to alternative DL- and non-DL-based methods. Models were tested on internal (n = 20) and external (n = 18) datasets from Latin America, and on an external dataset from Europe (n = 49). We also examined robustness by rescanning six patients (n = 6) from our MS clinical cohort. Moreover, we studied inter-human annotator agreement and discussed our findings in light of these results. Performance and robustness were assessed using intraclass correlation coefficient (ICC), Dice coefficient (DC), and coefficient of variation (CV). RESULTS: Inter-human ICC ranged from 0.89 to 0.95, while spatial agreement among annotators showed a median DC of 0.63. Using expert manual segmentations as ground truth, our DL model achieved a median DC of 0.73 on the internal, 0.66 on the external, and 0.70 on the challenge datasets. The performance of our DL model exceeded that of the alternative algorithms on all datasets. In the robustness experiment, our DL model also achieved higher DC (ranging from 0.82 to 0.90) and lower CV (ranging from 0.7 to 7.9%) when compared to the alternative methods. CONCLUSION: Our DL-based model outperformed alternative methods for brain MS lesion segmentation. The model also proved to generalize well on unseen data and has a robust performance and low processing times both on real-world and challenge-based data. CLINICAL RELEVANCE STATEMENT: Our DL-based model demonstrated superior performance in accurately segmenting brain MS lesions compared to alternative methods, indicating its potential for clinical application with improved accuracy, robustness, and efficiency. KEY POINTS: • Automated lesion load quantification in MS patients is valuable; however, more accurate methods are still necessary. • A novel deep learning model outperformed alternative MS lesion segmentation methods on multisite datasets. • Deep learning models are particularly suitable for MS lesion segmentation in clinical scenarios.

Biomarcadores radiológicos en RM para una aproximación al diagnóstico molecular en gliomas IDH-mutados (grado II y III) / MR Imaging Biomarkers for a Diagnostic Approach of IDH-mutated Gliomas (Grades II/III)

Resumen Objetivo: Analizar características por resonancia magnética (RM) de gliomas IDH-mutados (grado II y III) en base a parámetros cualitativos, a fin de valorar el rendimiento del signo del mismatch T2-FLAIR y otras características morfológicas de los tumores, en predecir el estado del 1p/19q y su reproducibilidad interobservador. Métodos Estudio retrospectivo, descriptivo y analítico sobre una cohorte de 53 gliomas IDH-mutados (grado II y III) y molecularmente definidos respecto al 1p/19q, seleccionados a partir de la base de datos de la institución, durante el periodo 2014- 2019. Dos neuroradiólogos evaluaron características imagenológicas de forma independiente y enmascarada al diagnóstico: mismatch T2-FLAIR, localización tumoral, bordes, señal, infiltración cortical e inhomogeneidad en T2. Los casos discordantes fueron evaluados por un tercer neuroradiólogo de mayor experiencia. Resultados: Treinta de 53 (56,6%) gliomas fueron no codelecionados, y 23/53 (43,4%) codelecionados. El signo del mismatch T2-FLAIR fue positivo en 16/53 (30,18%) pacientes, 15/16 (93,75%) no codelecionados y 1/16 (6,25%) codelecionado (Exacto de Fisher p = <,0001). Los dos evaluadores demostraron una concordancia interobservador casi perfecta para ese signo, κ =,907 (95% CI, 0,781 a 1,0). La especificidad y el valor predictivo positivo del signo para predecir la ausencia de la codeleción fue de un 95,7% y un 93,8% respectivamente. Discusión: La reciente actualización en la clasificación de los gliomas los clasifica acorde a su perfil molecular. En los últimos años, varios investigadores han estudiado características morfológicas por RM de los tumores con la intención de predecir las características moleculares de los mismos. Conclusión: En nuestra población, el signo del mismatch T2-FLAIR es el único biomarcador radiológico que muestra asociación estadísticamente significativa en predecir la ausencia de codeleción en los gliomas IDH-mutados (grado II y III), con una alta especificidad y un alto valor predictivo positivo.

Abstract Objective: To analyze magnetic resonance (MR) characteristics of IDH-mutated gliomas (grades II/III) utilizing qualitative parameters with the goal of assessing the performance of the T2-FLAIR mismatch sign and other morphological characteristics of tumors in predicting the 1p/19q co-deletion status as well as inter-observer reproducibility. Methods: Retrospective and descriptive study analyzing a cohort of 53 IDH-mutated lower-grade (grades II/III) gliomas with known 1p/19q co-deletion status. Patients meeting selection criteria for this study were taken from our institutional data from 2014-2019. Two neuroradiologists assessed the following imaging characteristics independently, and blinded from the diagnosis: T2-FLAIR mismatch, tumor location, borders, signal characteristics, cortical infiltration and T2* inhomogeneity. In the event of discordant interpretations, a third senior neuroradiologist also evaluated the case. Results: 23 of the 53 (43.4%) gliomas demonstrated 1p/19q co-deletion and 30 of 53 (56.6%) did not. T2-FLAIR mismatch was positive in 16 of 53 cases (30.2%) with 15 of 16 (93.8%) demonstrating no co-deletion and 1/16 (6.25%) with co-deletion (Fisher's exact p = < .0001). The two readers showed an almost perfect interreader agreement for this sign κ = 0.907 (95% CI, 0.781 to 1.0). Specificity and positive predictive value of the sign to predict the absence of co-deletion was 95.7% and 93.8% respectively. Discussion: The recent update in classification of lower-grade gliomas segregates gliomas according to molecular profile. In the recent past, many researchers have studied MR morphologic characteristics of these tumors with the intention of predicting molecular features of said tumors Conclusion: In our patient population, T2-FLAIR mismatch sign is the only radiologic biomarker that shows statistically significant association with the absence of 1p/19q co-deletion in lower-grade gliomas, with high specificity and positive predictive value.