portable constant-current electrical stimulator system compex motion ii  (Compex Inc)

Journal: medRxiv

Article Title: Fast intraoperative detection of primary CNS lymphoma and differentiation from common CNS tumors using stimulated Raman histology and deep learning

doi: 10.1101/2024.08.25.24312509

Figure Lengend Snippet: This figure demonstrates the intraoperative workflow and model implementation for detecting primary CNS lymphoma (PCNSL) within minutes using RapidLymphoma. On the left side, the workflow begins with extracting a fresh, unprocessed specimen from a patient with suspected PCNSL during a stereotactic-guided brain biopsy. This specimen is conventionally analyzed using H&E stained, FFPE-based analysis, and an intraoperative Stimulated Raman Histology (SRH) imager. After squeezing on a slide, the SRH imager generates a high-resolution SRH image, highlighting histopathological features. This image is then subjected to patching. Each patch is processed by a segmentation model for diagnostic (tumor, normal brain tissue) and non-diagnostic prediction and is visualized as a heatmap overlay, where non-diagnostic regions are marked in blue and diagnostic tumor regions in red. If tumor regions are detected, our self-supervised vision encoder (BYOL-optimized), called RapidLymphoma, generates latent representations (Z 1 , Z 2 , …, Z n-1 , Z n ) of these diagnostic patches. These latent representations are fed into a classification head for our final downstream task to detect PCNSL and differentiate from non-PCNSL brain tumors. The corresponding heatmap overlay excludes non-diagnostic areas (grey) from diagnostic areas (Blue: non-PCNSL, red: PCNSL), with high slide-averaged confidence percentages. Inset images of representative patches illustrate the cellular details corresponding to low-confidence and high-confidence areas for PCNSL, visualizing the model’s predictions.

Article Snippet: The whole-slide images used to develop the AI-based pipeline and the prospective international multicenter clinical trial were obtained using a portable fiber-laser-based stimulated Raman scattering (SRS) microscope (NIO Laser Imaging System, Invenio Imaging Inc., Santa Clara, CA, USA).

Techniques: Staining, Diagnostic Assay

Journal: medRxiv

Article Title: Fast intraoperative detection of primary CNS lymphoma and differentiation from common CNS tumors using stimulated Raman histology and deep learning

doi: 10.1101/2024.08.25.24312509

Figure Lengend Snippet: Correctly predicted whole slide images from the prospective testing cohort demonstrate the model’s effectiveness in distinguishing between PCNSL and non-PCNSL tissues. The figure is organized into three main columns: Stimulated Raman Histology (SRH) images, RapidLymphoma’s prediction heatmap overlays, and detailed image patch comparisons. Red indicates high confidence for PCNSL, blue indicates non-PCNSL, and white/grey indicates low confidence. The detailed image patches on the right side from the first two PCNSL slides demonstrate an angiogenic growth pattern, high nuclear/cytoplasmic ratio, monomorphic cell structure, and reticulin pattern. In contrast, patches from the other two non-PCNSL slides illustrate polymorphic cell structure, lower nuclear/cytoplasmic ratio, and astrocytic cell structure. The first two rows are featured by a diffuse large B-cell lymphoma, the other two by a plasmocytoma, and an astrocytoma, IDH-mutant CNS WHO grade 4. Image quality impacts RapidLymphoma’s predictions with lower confidence, resulting in white/grey patch regions.

Article Snippet: The whole-slide images used to develop the AI-based pipeline and the prospective international multicenter clinical trial were obtained using a portable fiber-laser-based stimulated Raman scattering (SRS) microscope (NIO Laser Imaging System, Invenio Imaging Inc., Santa Clara, CA, USA).

Techniques: Mutagenesis

Journal: medRxiv

Article Title: Fast intraoperative detection of primary CNS lymphoma and differentiation from common CNS tumors using stimulated Raman histology and deep learning

doi: 10.1101/2024.08.25.24312509

Figure Lengend Snippet: This figure displays four false positive whole slide images from the prospective test cohort, illustrating the challenges and complexities in distinguishing between PCNSL and non-PCNSL tissues. On the left the Stimulated Raman Histology (SRH) whole slide images, RapidLymphoma’s prediction heatmap overlays in the middle, and detailed patch comparisons from the marked red and blue squares indicating the model’s predicted regions as PCNSL and non-PCNSL, respectively on the right. For each misclassified case, the actual diagnosis and the model’s incorrect prediction are noted alongside the softmax scores on slide-level. The heatmaps use color gradients to represent the model’s confidence levels, with red indicating high confidence for PCNSL, blue for non-PCNSL, and white/grey for low confidence. A detailed analysis of the patches from the misclassified slides shows similar features across the two classes, e.g., high nuclear/cytoplasmic ratio or monomorphic cell patterns. Noisy images, such as from small biopsy tissue samples (Image 3 and 4), also influence prediction behavior.

Article Snippet: The whole-slide images used to develop the AI-based pipeline and the prospective international multicenter clinical trial were obtained using a portable fiber-laser-based stimulated Raman scattering (SRS) microscope (NIO Laser Imaging System, Invenio Imaging Inc., Santa Clara, CA, USA).

Techniques: Biomarker Discovery

Journal: medRxiv

Article Title: Fast intraoperative detection of primary CNS lymphoma and differentiation from common CNS tumors using stimulated Raman histology and deep learning

doi: 10.1101/2024.08.25.24312509

Figure Lengend Snippet: The figure illustrates the performance of RapidLymphoma in distinguishing primary central nervous system lymphoma (PCNSL) from two common differential diagnoses: adult-type diffuse glioma (IDH-wildtype) and metastasis. The data was derived from independent and external test cohorts to ensure further robust evaluation. On the left side, performance metrics are presented for HGG, and on the right, for metastasis. Due to the rarity and limited availability of lymphoma cases, the same stimulated Raman histology (SRH) images from the prospective cohort were utilized to generate the metrics. These presented test cohorts are independent and external from the primary prospective test cohort. This approach was adopted to provide a more accurate depiction of the classifier’s performance against the typical differential diagnoses such as IDH-wildtype glioma (Glioblastoma, diffuse midline glioma, gliosarcoma, and diffuse hemispheric glioma) and metastasis (Melanoma, various adenocarcinoma, squamous cell carcinoma, mamma carcinoma, and neuroendocrine carcinoma). A. A Receiver Operating Characteristic (ROC) curve represents the classifier’s ability to differentiate between PCNSL and adult-type diffuse glioma (IDH-wildtype), indicating high diagnostic performance. B. This plot demonstrates the classifier’s performance in distinguishing PCNSL from brain metastasis with another ROC curve, suggesting excellent classification capability. C. This panel shows the performance metrics for PCNSL versus adult-type diffuse glioma (IDH-wildtype). The bar charts present the balanced accuracy, specificity, and sensitivity at the patient level (n = 437) and slide level (n = 1736). The classifier exhibits high performance across all metrics, indicating its reliability in distinguishing PCNSL from high-grade glioma. D. The performance metrics for PCNSL versus metastasis are depicted in this panel. Like panel C, these bar charts present balanced accuracy, specificity, and sensitivity at the patient level (n = 59) and slide level (n = 149). The results consistently show high performance, underscoring the classifier’s effectiveness in differentiating PCNSL from brain metastasis. E. The confusion matrices for PCNSL versus adult-type diffuse glioma (IDH-wildtype; HGG) are shown in this panel to reveal false positive and negative predictions. F. The confusion matrices for PCNSL versus metastasis are illustrated in this panel.

Article Snippet: The whole-slide images used to develop the AI-based pipeline and the prospective international multicenter clinical trial were obtained using a portable fiber-laser-based stimulated Raman scattering (SRS) microscope (NIO Laser Imaging System, Invenio Imaging Inc., Santa Clara, CA, USA).

Techniques: Derivative Assay, Diagnostic Assay