Purpose
-
Introduction of specialized DL models training on SSeq-MRI and MS-MRI.
-
To optimally select SS-MRI for timely intervention in CP identification from infancy to adolescence.
-
Benchmarking of MS-MRI coupling to reduce CP misclassification employing contrastive learning.
-
Visualizing CP’s vulnerable regions in MRI scans and slices to assist radiomics.
-
Delving into the trends associated with HC and CP from infancy to adolescence.
Material and methods
Single scan MRI-based modeling
Multi-contrast learning and CP identification
Dataset collection
MRI machine | HC | Patient | PSNR | Contrast between tissues | Information potential | Anatomical details |
---|---|---|---|---|---|---|
Skyra (3T) | 123 | 44 | High | High | High | High |
GE (1.5) | 138 | 282 | Low | Low | Low | Low |
Philips (3T) | 128 | 1 | High | High | High | High |
Total | 389 | 327 |
Subject | G_1 | G1_3 | G3_5 | G5_8 | G8_11 | G11_13 | G13_15 | G15_18 |
---|---|---|---|---|---|---|---|---|
Age range | <1 | >1:\(<=\)2 | >2:\(<=\)4 | >4:\(<=\)7 | >7:\(<=\)10 | <10:\(<=\)13 | >13:\(<=\)15 | > 15 |
Controls | 16 | 68 | 59 | 101 | 82 | 49 | 10 | 3 |
CP | 52 | 95 | 90 | 54 | 13 | 10 | 3 | 6 |
Subject | Criteria | Description | ||||||
---|---|---|---|---|---|---|---|---|
CP | Inclusion | 1. CP confirmation via international consensus criteria. | ||||||
2. Complete routine MRI examination. | ||||||||
Exclusion | 1. Having other neurological disorders, traumatic brain injuries, and systemic illnesses. | |||||||
2. Artifacts and poor image quality. | ||||||||
Controls | Inclusion | 1. Apgar score >=8 at 1 min and 7 min after birth. Full term and of normal weight. | ||||||
2. No abnormality. | ||||||||
Exclusion | 1. Perinatal asphyxia, intrauterine distress, or any neurological disease, traumatic brain injuries, or systemic illnesses. | |||||||
2. Artifacts and poor image quality. |
Clinical characteristics of patients
Characteristics | CP | HC | CP and HC | P value |
---|---|---|---|---|
Age: Mean [s.d] | 3.446 [3.227] | 6.280 [3.781] | 4.863 [3.811] | <0.001*** |
Gender: N [Male/Female] | 327 [206/121] | 381 [233/148] | – | – |
Weight: Mean [s.d] | 7.523 [8.335] | 14.365 [13.970] | 5.628 [13.582] | <0.001*** |
CP Type | ||||
CP (not classified): N [%] | 98 [29.96%] | – | – | – |
Spastic: N [%] | 192 [58.71%] | – | – | – |
Dyskinetic: N [%] | 18 [5.50%] | – | – | – |
Ataxic: N [%] | 0 [0%] | – | – | – |
Hypotonic: N [%] | 0 [0%] | – | – | – |
Mixed: N [%] | 19 [5.81%] | – | – | – |
Rigid: N [%] | 0 [0%] | – | – | – |
Paralysis Type | – | – | – | – |
Mixed/Unknown: N [%] | 127 [38.83%] | – | – | – |
Monoplegia: N [%] | 3 [0.91] | – | – | – |
Diplegia: N [%] | 59 [18.04%] | – | – | – |
Triplegia: N [%] | 0 [0%] | – | – | – |
Hemiplegia: N [%] | 62 [18.96%] | – | – | – |
Tetraplegia: N [%] | 76 [23.24%] | – | – | – |
CP term: Preterm N [%], Full term N [%] | 155 [47.40%], 172 [52.59%] | – | – | – |
Delivery: Cesarean N [%], Normal [%] | 146 [44.64%], 181 [55.35%] | – | – | – |
Movement Disorder: Yes N [%], No N [%] | 274 [83.79%], 53 [16.20%] | – | – | – |
Mental retardation: Yes N [%], No N [%] | 156 [47.70%], 171 [52.29%] | – | – | – |
Swallowing dysfunction: Yes N [%], No N [%] | 42 [12.84%], 285 [87.15%] | – | – | – |
Speech disorder: Yes N [%], No N [%] | 114 [34.86%], 213 [65.13%] | – | – | – |
Epilepsy: Yes N [%], No N [%] | 97 [29.66%], 230 [70.33%] | – | – | – |
Rehabilitation received: Yes N [%], No N [%] | 253 [77.37%], 74 [22.62%] | – | – | – |
Motor delay function: Yes N [%], No N [%] | 267 [81.65%], 60 [18.34%] | – | – | – |
Traction: Yes N [%], No N [%] | 5 [1.52%], 322 [98.47%] | – | – | – |
Balance function training: Yes N [%], No N [%] | 8 [2.44%], 319 [97.55%] | – | – | – |
Hyperbaric oxygen chamber therapy: Yes N [%], No N [%] | 52 [15.90%], 275 [84.09%] | – | – | – |
Direct current therapy: Yes N [%], No N [%] | 114 [34.86%], 213 [65.13%] | – | – | – |
Botulinum toxin injection: Yes N [%], No N [%] | 26 [7.95%], 301 [92.04%] | – | – | – |
Wearing aligner: Yes N [%], No N [%] | 22 [6.72%], 305 [93.27%] | – | – | – |
Cord blood transplantation treatment: Yes N [%], No N [%] | 14 [4.28%], 313 [95.71%] | – | – | – |
MRI slices distribution and covered brain regions
Results
Single scan (SC) brain MRI and CP association
Scan | Model | TP | FP | TN | FN | Specificity | Sensitivity | PPV | NPV | F\(_{1}\) | Accuracy (%) |
---|---|---|---|---|---|---|---|---|---|---|---|
T2-w | SSeq-DL | 56 | 4 | 35 | 7 | 0.8974 | 0.8889 | 0.9334 | 0.8334 | 0.8931 | 89.21 |
Flair | SSeq-DL | 55 | 5 | 36 | 6 | 0.8780 | 0.9016 | 0.9166 | 0.8571 | 0.8896 | 89.21 |
T1-w | SSeq-DL | 53 | 7 | 37 | 5 | 0.8409 | 0.9137 | 0.8833 | 0.8809 | 0.8758 | 88.23 |
Sag | SSeq-DL | 53 | 7 | 37 | 5 | 0.8409 | 0.9137 | 0.8833 | 0.8809 | 0.8758 | 88.23 |
Enhancing CP identification using MS-MRI-based learning
Fused Models | Model | TP | FP | TN | FN | Specificity | Sensitivity | PPV | NPV | F\(_1\) | Acc |
---|---|---|---|---|---|---|---|---|---|---|---|
CrossValidation-1 | 57 | 3 | 38 | 4 | 0.926 | 0.9344 | 0.9500 | 0.9047 | 0.9306 | 93.13 | |
CrossValidation-2 | 56 | 4 | 38 | 4 | 0.9047 | 0.9333 | 0.9333 | 0.9047 | 0.9188 | 92.15 | |
Fusion of T1-w and Flair | CrossValidation-3 | 57 | 3 | 41 | 1 | 0.9318 | 0.9827 | 0.9500 | 0.9761 | 0.9566 | 96.07 |
CrossValidation-4 | 57 | 3 | 40 | 2 | 0.9302 | 0.9661 | 0.9500 | 0.9523 | 0.9478 | 95.09 | |
CrossValidation-5 | 58 | 2 | 41 | 1 | 0.9534 | 0.9830 | 0.9666 | 0.9761 | 0.9680 | 97.05 | |
Cumulative scores | 0.9294 | 0.9599 | 0.9500 | 0.9428 | 0.9443 | 94.70 | |||||
CrossValidation-1 | 57 | 3 | 40 | 2 | 0.93023 | 0.96610 | 0.9500 | 0.9723 | 0.9478 | 95.09 | |
CrossValidation-2 | 56 | 4 | 38 | 4 | 0.9047 | 0.9333 | 0.9333 | 0.9523 | 0.9047 | 92.15 | |
Fusion of T2-w and Flair | CrossValidation-3 | 55 | 5 | 40 | 2 | 0.8936 | 1 | 0.9166 | 1 | 0.9438 | 95.09 |
CrossValidation-4 | 59 | 1 | 40 | 2 | 0.9756 | 0.9642 | 0.9833 | 0.9523 | 0.9797 | 98.03 | |
CrossValidation-5 | 54 | 6 | 40 | 2 | 0.8695 | 0.9642 | 0.9000 | 0.9523 | 0.9144 | 92.15 | |
Cumulative scores | 0.9147 | 0.9661 | 0.9366 | 0.9523 | 0.9392 | 94.31 |