A convolutional neural network (CNN) is proposed to learn multiple useful feature representations for a classification from low level (raw pixels) to high level (object). Convolutional kernels are initialized by the learned filter kernels that come from sparse auto-encoders. Unlike some traditional methods, which divide the feature abstracting and classifier training into two separated processes, a discriminative feature vector and a single multi-class classifier of softmax regression are learned simultaneously during the training process. Based on the learned high-quality feature representation, the classification can be efficiently performed. A real-world case of surface defects on steel sheet, which evaluates the classification performance of the proposed method, is depicted in detail. The experimental results indicate that the proposed method is quite simple, effective and robustness for the classification of surface defects on hot-rolled steel sheet. Keywords: convolutional neural networks, classification, surface defects, steel sheet, convolutional kernels, sparse auto-encoder Konvolucijska nevronska mre`a (CNN) je predlagana za u~enje {tevilnih koristnih predstavitev pri klasifikaciji od nizkega nivoja (grobe slikovne pike) do visokega nivoja (predmet). Konvolucijska jedra so inicializirana z nau~enimi filtrirnimi jedri, ki izhajajo iz redkih samoenkoderjev. Razli~no od nekaterih klasi~nih metod, ki delijo funkcijo abstrakcije in trening klasifikacije v dva lo~ena procesa, se vektor nediskriminativne funkcije v enostavnem ve~razrednem klasifikatorju regresije softmax, u~i hkrati med procesom treninga. Na osnovi nau~ene predstavitve z visoko kvalitetno funkcijo, se lahko klasifikacija u~inkovito izvede. Primer iz resni~nega sveta povr{inske napake na jekleni plo~evini, ki ocenjujejo zmogljivost klasifikacije je prikazan v podrobnostih. Rezultati eksperimentov ka`ejo, da je predlagana metoda razmeroma preprosta, u~inkovita in robustna pri klasifikaciji povr{inskih napak na vro~e valjani jekleni plo~evini.

Object The surgical treatment of severe and rigid spinal deformities poses difficulties and dangers. In this article, the authors summarize their surgical techniques and evaluate patient outcomes after performing posterior vertebral column resection (PVCR) for the correction of spinal deformities with curves greater than 100°, and investigate the crucial points to ensure neurological safety during this challenging procedure. Methods The authors retrospectively reviewed their experience with 28 patients with extremely severe (Cobb angles in the coronal or sagittal plane > 100°) and rigid thoracic or thoracolumbar spine deformities who underwent PVCR. The average patient age was 20.2 years and all patients underwent a minimum follow-up of 24 months (range 24–60 months). Patients were divided into groups according to their morphological classification as follows: kyphosis alone (Group A, 6 patients with a mean Cobb angle of 109.0° [range 105°–120°]); kyphoscoliosis with coronal plane curves notably greater than sagittal plane curves (Group B, 14 patients with mean scoliotic curves of 116.6° [range 102°–170°] and kyphotic curves of 77.7° [range 42°–160°]); and kyphoscoliosis with sagittal curves notably greater than coronal plane curves (Group C, 8 patients with a mean coronal curve of 85.4° [range 65°–110°] and a mean sagittal curve of 117.6° [range 102°–155°]). Results A total of 36 vertebrae were removed in 28 patients who had a severe rigid spinal deformity, and the mean fusion extent was 13.3 vertebrae (range 7–17 vertebrae). The mean operating time was 620 minutes (range 320–920 minutes) with an average operative blood loss of 6,680 ml (range 3,000–24,000 ml). The overall final correction rate of scoliosis was 59.0%, and average postoperative kyphotic Cobb angles ranged from 30.4° to 95.9°. In Group A the mean preoperative sagittal angle of 109.0° was corrected to a mean postoperative angle of 32.0°. In the Group B kyphoscoliotic patients, the correction rate in the coronal plane was 58.6%; the Cobb angle in the sagittal plane was corrected from a mean of 77.7° preoperatively to 25.1° postoperatively; in Group C, the correction rate in the coronal plane was 58.5%, and the mean sagittal angle was reduced from a mean of 117.6° preoperatively to 39.0°. Of the 28 patients who underwent PVCR, 46 complications were observed in 18 patients intra- and postoperatively. There were 5 neurological complications including 1 case of late-onset paralysis and 4 cases of thoracic nerve root pain, all of which resolved during the early follow-up period. Nonneurological complications occurred more often in kyphoscoliotic patients (41 complications). The mean follow-up of all patients was 33.7 months (range 24–60 months). Conclusions Posterior vertebral column resection was effective in correcting severe rigid spinal deformity, although the procedure was technically demanding, exhaustingly lengthy, and was associated with a variety of complications. The PVCR technique created a space for spinal correction and spinal cord tension adjustment and the correction could be performed under direct inspection and by palpation of the tension in the spinal cord through the space. Therefore, in terms of the spinal cord, the deformity correction process involved in the PVCR procedure is relatively safe.

Purpose To determine the risk factors of neurologic deficits during PVCR correction, so as to help improve safety during and after surgery. Methods A consecutive series of 76 patients with severe and rigid spinal deformities who were treated with PVCR at a single institution between October 2004 and July 2011 were included in our study. Of the 76 patients, 37 were male and 39 female, with an average age of 17.5 years (range 10-48 years). There were 52 adolescent patients (with an age \18 years) and 24 adult patients (with an age C18 years). Preoperatively, postoperatively and 6 months after surgery, we performed systemically neurologic function evaluations of each patients through meticulous physical examination. Any new abnormality or deterioration in evaluation of neurologic function than preoperative is reckoned postoperative neurologic deficits. Ten variables that might affect the safety of neurologic deficits during PVCR procedures, including imaging factors, clinical factors and operational factors, were analyzed using univariate analysis. Then the variables with statistical difference were analyzed by using multi-factor unconditional logistic regression analysis. Results No patient in this series had permanent paraplegia and nerve root injury due to operation. Change of neurologic status was found in six patients after surgery. Results of single-factor comparison demonstrated that the following seven variables were statistically different (P \ 0.05): location of apex at main curve (X 3 ), Cobb angle at the main curve at the coronal plane (X 4 ), scoliosis associated with thoracic hyperkyphosis (X 5 ), level of vertebral column resected (X 6 ), number of segmental vessels ligated (X 7 ), preexisting neurologic dysfunction (X 8 ), and associated with intraspinal and brain stem anomalies (X 9 ). The multi-factor unconditional logistic regression analysis revealed that X 8 (OR = 49.322), X 9 (OR = 18.423), X 5 (OR = 11.883), and X 6 (OR = 8.769) were independent and positively correlated with the neurologic deficit. Conclusions Preexisting neurologic dysfunction, associated with intraspinal and brain stem anomalies, scoliosis associated with thoracic hyperkyphosis and level of vertebral column resected are independent risk factors for neurologic deficits during PVCR procedure.

No abstract

Priority of neurological decompression was regarded as necessary for scoliosis patients associated with Chiari I malformation in order to decrease the risk of spinal cord injury from scoliosis surgery. We report a retrospective series of scoliosis associated with Chiari I malformation in 13 adolescent patients and explore the effectiveness and safety of posterior scoliosis correction without suboccipital decompression. One-stage posterior approach total vertebral column resection was performed in seven patients with scoliosis or kyphosis curve [90°(average 100.1°s coliotic and 97.1°kyphotic curves) or presented with apparent neurological deficits, whereas the other six patients underwent posterior pedicle screw instrumentation for correction of spinal deformity alone (average 77.3°s coliotic and 44.0°kyphotic curves). The apex of the scoliosis curve was located at T7-T12. Mean operating time and intraoperative hemorrhage was 463 min and 5,190 ml in patients undergoing total vertebral column resection, with average correction rate of scoliosis and kyphosis being 63.3 and 71.1%, respectively. Mean operating time and intraoperative hemorrhage in patients undergoing instrumentation alone was 246 min and 1,450 ml, with the average correction rate of scoliosis and kyphosis being 60.8 and 53.4%, respectively. The mean follow-up duration was 32.2 months. No iatrogenic neurological deterioration had been encountered during the operation procedure and follow-up. After vertebral column resection, neurological dysfunctions such as relaxation of anal sphincter or hypermyotonia that occurred in three patients preoperatively improved gradually. In summary, suboccipital decompression prior to correction of spine deformity may not always be necessary for adolescent patients with scoliosis associated with Chiari I malformation. Particularly in patients with a severe and rigid curve or with significant neurological deficits, posterior approach total vertebral column resection is likely a good option, which could not only result in satisfactory correction of deformity, but also decrease the risk of neurological injury secondary to surgical intervention by shortening spine and reducing the tension of spinal cord.