f1 score  (Thermo Fisher)

Journal: Cell Reports Methods

Article Title: A deep learning pipeline for accurate and automated restoration, segmentation, and quantification of dendritic spines

doi: 10.1016/j.crmeth.2025.101179

Figure Lengend Snippet: Importance of image restoration in spine detection accuracy (A–C) Maximum intensity projection images of a dendritic segment acquired under low-SNR conditions (A), following CARE restoration (B), and at a high SNR (C). The right images in each row show RESPAN’s segmentation outputs, with color-coded spines denoting true positives (TPs; green), false positives (FPs; orange), or false negatives (FNs; magenta). Insets highlight spines that are barely distinguishable in the low-SNR dataset but can be resolved following restoration. (D) Detection rates (TPs, FPs, and FNs) for low-SNR (black) vs. restored (magenta) datasets. Statistical comparisons were performed using Wilcoxon matched-pairs signed-rank tests with a two-stage step-up method to control the false discovery rate. (E) Spine mask intersection-over-union (IoU = 0.5) analysis comparing segmentation labels from low-SNR and restored images to those from the high-SNR output. Each point represents a matched spine, and the line/whiskers depict the median and 95% confidence interval (CI); p values were obtained using Wilcoxon matched-pairs signed-rank tests. (F) F1 scores plotted across increasing IoU thresholds (0.1–0.9). Comparisons between low-SNR (black) and restored (magenta) curves were performed using Wilcoxon matched-pairs signed-rank tests. (G) Cumulative distributions of Hausdorff distances for low-SNR (black) and restored (magenta) spines, where lower values indicate closer alignment with the ground-truth spine shape. Data were analyzed by a Kolmogorov-Smirnov test, revealing significantly reduced Hausdorff distances in restored data.

Article Snippet: Quantitative metrics confirm that RESPAN achieves superior recall ( p = 0.0144 vs. DeepD3; p = 0.0954 vs. Imaris), precision ( p = 0.0143 vs. DeepD3; p = 0.0177 vs. Imaris), and F1 score ( p = 0.0047 vs. DeepD3; p = 0.0473 vs. Imaris) ( H).

Techniques: Control

Journal: Cell Reports Methods

Article Title: A deep learning pipeline for accurate and automated restoration, segmentation, and quantification of dendritic spines

doi: 10.1016/j.crmeth.2025.101179

Figure Lengend Snippet: Comparison of RESPAN performance with other software (A) Maximum intensity projection of a raw fluorescence dataset showing a dendritic segment. (B–D) Spine detection outputs from RESPAN (B), DeepD3 (C), and Imaris (D) overlaying the dendritic segment. Detected spines are color coded as true positives (TPs; green), false positives (FPs; magenta), and false negatives (FNs; orange). (E) Percentage of TP spines detected by each method. (F) Recall scores for spine detection, with RESPAN demonstrating consistently higher recall, reflecting fewer FNs. (G) Precision scores for spine detection, with RESPAN outperforming DeepD3 and Imaris by reducing FP detections. (H) F1 scores for spine detection, representing the harmonic mean of precision and recall. RESPAN achieves superior F1 scores compared to other methods. (E)–(H) show a solid line at the median. Sample size: 440 GT spines from 11 dendritic segments.

Article Snippet: Quantitative metrics confirm that RESPAN achieves superior recall ( p = 0.0144 vs. DeepD3; p = 0.0954 vs. Imaris), precision ( p = 0.0143 vs. DeepD3; p = 0.0177 vs. Imaris), and F1 score ( p = 0.0047 vs. DeepD3; p = 0.0473 vs. Imaris) ( H).

Techniques: Comparison, Software, Fluorescence

Journal: Diagnostics

Article Title: Detection of Fractured Endodontic Instruments in Periapical Radiographs: A Comparative Study of YOLOv8 and Mask R-CNN

doi: 10.3390/diagnostics15060653

Figure Lengend Snippet: Inter- and Intra-group Comparisons of F1 Scores for Detection of FEI, RCT, and BG by Endodontist A, Endodontist B, Mask R-CNN, and YOLOv8.

Article Snippet: FEI F1 Score , 0.9947 , 0.9947 , 0.9890 , 1.0000 , 1.000 , 0.640 , 0.447 , 0.640 , 0.447 , 0.272.

Techniques: