Article Figures & Data
Figures
- FIG 1.
Exemplary axial images from T2-weighted (A) and FLAIR (B) sequences of an included patient.
- FIG 2.
Our AI pipeline for training an outcome prediction model on shunted NPH baseline MRI data set (A), including a detailed review of image pre-processing (B).
- FIG 3.
Structure of ResNet50 network.
- FIG 4.
Fusion prediction model with ResNet50 models for integrating information from multiple imaging sequences.
- FIG 5.
Receiver operating characteristic (ROC) curves demonstrating comparison between models trained on T2-weighted, FLAIR, and both sequences for each of 4 clinical improvement end points on primary institution testing data set.
Tables
Patient Characteristic Mean (± SD) or Median (IQR) Training/Testing (Institution 1, n = 249) Validation (Institution 2, n = 33) Age 74.2 (± 7.5) years 72.1 (± 7.7) years Sex Men 129 (51.8%) 19 (57.6%) Women 120 (48.2%) 14 (42.4%) Race White 232 (93.2%) 28 (84.8%) Black/African American 7 (2.8%) 4 (12.1%) Other/unknown 10 (4.0%) 1 (3.0%) BMI 28.5 (± 5.8) 29.1 (±4.6) CCI 6 (5–7) 4 (3–6) Shunt laterality Right 239 (95.9%) 33 (100%) Left 10 (4.1%) 0 (0%) Catheter terminus placement Ventriculoperitoneal 244 (98.0%) 33 (100%) Ventriculopleural 5 (2.0%) 0 NPH classification iNPH 212 (85.1%) 32 (97.0%) sNPH 37 (14.9%) 1 (3.0%) Baseline mRS 0– No symptoms 0 0 1– No significant disability 1 (0.4%) 0 2– Slight disability 107 (43.0%) 8 (24.2%) 3– Moderate disability 83 (33.3%) 8 (24.2%) 4– Moderate-severe disability 41 (16.5%) 15 (45.5%) 5– Severe disability 3 (1.2%) 1 (3.0%) Unable to determine 14 (5.6%) 0 8-Point gait score 4 (4–6) 5 (4–6) 6-Point incontinence score 3 (2–4) 3 (2–4) Note:—BMI indicates body mass index; SD, standard deviation.
- Table 2:
Presenting symptoms and postoperative improvement; denominators for improvement calculations are patients who had each symptom at baseline and attended follow-up
Triad of NPH Symptoms and Functional Disability Present at Baseline (Pre-Op) Improved at 3 Months Post-Op Improved at 12 Months Post-Op Training & testing cohort 249/249 (100%) 226/249 (90.8%) 175/249 (70.3%) Gait impairment 249 (100%) 165/226 (73.0%) 123/175 (70.3%) Urinary incontinence 196 (78.7%) 76/127 (59.8%) 66/97 (68.0%) Cognitive impairment 217 (87.1%) 97/208 (46.6%) 75/160 (46.9%) mRS –– 105/179 (58.7%) 84/149 (56.4%) Validation cohort 33/33 (100%) 33/33 (100%) 33/33 (100%) Gait impairment 33 (100%) 29/33 (87.9%) 22/33 (66.7%) Urinary incontinence 21 (63.6%) 4/33 (12.1%) 3/33 (9.1%) Cognitive impairment 31 (93.9%) 14/33 (42.4%) 12/33 (36.4%) mRS –– 7/33 (21.2%) 6/33 (18.2%) Model Input mRS Improvement Gait Improvement Urinary Improvement Cognitive Improvement Age, sex, CCI 0.6255 [0.5089–0.7421] 0.5413 [0.4205–0.6621] 0.7135 [0.5306–0.8964] 0.6562 [0.4918–0.8207] T2 only 0.5125 [0.3392–0.6858] 0.6994 [0.5901–0.8087] 0.7304 [0.6253–0.8355] 0.4479 [0.2763–0.6195] FLAIR only 0.6723 [0.4968–0.8478] 0.7195 [0.6046–0.8344] 0.7929 [0.6912–0.8947] 0.7175 [0.5533–0.8816] T2+FLAIR 0.7395 [0.5765–0.9024] 0.8816 [0.8030–0.9602] 0.7874 [0.6845–0.8903] 0.7230 [0.5600–0.8859] Note:—AUROC values represent the results of pretrained models when evaluated on testing data set and 95% CI.
Model Input mRS Improvement Gait Improvement Urinary Improvement Cognitive Improvement T2 only 0.5513 [0.2874–0.8152] 0.6167 [0.2391–0.9942] 0.7816 [0.4609–1] 0.5732 [0.4326–0.7139] FLAIR only 0.7989 [0.6909–0.9068] 0.4758 [0.0517–0.8999] 0.6724 [0.3645–0.9803] 0.6364 [0.5024–0.7703] T2+FLAIR 0.8291 [0.7271–0.9311] 0.7333 [0.4263–1] 0.7586 [0.4286–1] 0.6310 [0.4326–0.8293] Note:—AUROC values represent the results of the same pretrained models when evaluated on external validation data set and 95% CI.
Jump to section
Related Articles
Cited By...
- No citing articles found.