Summary
Vestibular schwannomas (VS) are the most typical tumor of the cranium base with out there remedy choices that carry a danger of iatrogenic harm to the facial nerve, which might considerably influence sufferers’ high quality of life. As facial nerve outcomes stay difficult to prognosticate, we endeavored to make the most of machine studying to decipher predictive components related to facial nerve outcomes following microsurgical resection of VS. A database of patient-, tumor- and surgery-specific options was constructed through retrospective chart overview of 242 consecutive sufferers who underwent microsurgical resection of VS over a 7-year examine interval. This database was then used to coach non-linear supervised machine studying classifiers to foretell facial nerve preservation, outlined as Home-Brackmann (HB) I vs. facial nerve harm, outlined as HB II–VI, as decided at 6-month outpatient follow-up. A random forest algorithm demonstrated 90.5% accuracy, 90% sensitivity and 90% specificity in facial nerve harm prognostication. A random variable (rv) was generated by randomly sampling a Gaussian distribution and used as a benchmark to match the predictiveness of different options. This evaluation revealed age, physique mass index (BMI), case size and the tumor dimension representing tumor progress in direction of the brainstem as prognosticators of facial nerve harm. When validated through potential evaluation of facial nerve harm danger, this mannequin demonstrated 84% accuracy. Right here, we describe the event of a machine studying algorithm to foretell the chance of facial nerve harm following microsurgical resection of VS. Along with serving as a clinically relevant device, this highlights the potential of machine studying to disclose non-linear relationships between variables which can have scientific worth in prognostication of outcomes for high-risk surgical procedures.
Introduction
Vestibular schwannomas (VS; previously acoustic neuroma) are the most typical tumor of the cranium base with practically 2500 new circumstances identified within the US annually1 and accounting for 8% of all intracranial tumors2. VS proceed to current a scientific conundrum in that their benign pathology affords a sluggish progress sample with low chance of metastasis; nonetheless, native compression of cranial nerve VIII (the vestibulocochlear nerve) and the brainstem may end up in listening to loss, dizziness, vertigo and within the worst circumstances, sudden loss of life3. Moreover, microsurgical resection carries an inherent danger of iatrogenic harm to every of those constructions with comparable scientific penalties. Microsurgery carries an extra danger of harm to the close by cranial nerve VII (facial nerve) which can lead to important morbidity and impairment in high quality of life. As such, the chance of harm to every of those constructions from remedy have to be weighed towards the chance of creating problems from the pure historical past of tumor development.
Traditionally, remedy selections have been based mostly on tumor measurement and progress patterns over time. Remedy is commonly thought-about when serial progress is noticed on interval imaging, or sufferers develop neurological signs that correlate with tumor compression. Microsurgery stays the mainstay of remedy for giant VS4. For lesions > 2.5 cm, vestibulocochlear nerve compression past salvageability is commonly encountered and sufferers are preemptively recommended that listening to preservation is unlikely5. Nonetheless, facial nerve preservation stays an essential aim of surgical procedure. Whereas bigger tumors are typically related to harder facial nerve dissection, few different components that portend a better danger of facial nerve dysfunction have been recognized, and thus prognostication on the particular person affected person degree stays comparatively poor6. Even in conditions the place anatomic preservation of the nerve is achieved, stretching or different trauma from troublesome dissection can result in post-operative facial weak point. Whereas machine studying algorithms have not too long ago been developed within the area of VS to help in selections concerning timing of remedy7 and chance of listening to preservation8, no examine has but utilized this know-how to discern the chance of facial nerve dysfunction, nor to develop a deeper understanding of the clinically related components which can contribute to poorer facial nerve outcomes. We leveraged rising machine studying approaches mixed with the VS expertise at two high-volume VS facilities to develop an algorithm for prediction of facial nerve dysfunction in sufferers present process microsurgical resection of VS, based mostly on affected person, surgical procedure, and tumor traits.
Strategies
Database assortment
This examine was reviewed by the College of Pennsylvania Institutional Overview Board, who decided that it met standards for exemption from full moral approval and topic knowledgeable consent. All strategies had been carried out in accordance with related tips and rules. Information from the College of Pittsburgh had been shared in accordance with executed Information Utilization Agreements between the College of Pittsburgh and the College of Pennsylvania. We performed a retrospective chart overview of sufferers who had undergone retrosigmoid or translabyrinthine craniotomies for VS over a seven-year examine interval from 2014 to 2021 at three hospitals: the Hospital of the College of Pennsylvania, Pennsylvania Hospital, and the College of Pittsburgh Medical Middle Presbyterian Hospital. We excluded sufferers who underwent retrosigmoid or translabyrinthine craniotomies for different pathologies, together with decrease cranial nerve schwannomas, meningiomas, chordomas, epidermoid cysts and mind metastasis. Two sufferers expired previous to the 6-month follow-up interval and are thus not represented on this evaluation. Information reviewed included affected person demographic data, surgical studies, and pre- and post-operative magnetic resonance pictures.
Statistical strategies
The first final result assessed was facial nerve operate at 6-month follow-up. Facial nerve operate was assessed on the idea of doctor rankings of facial operate, measured utilizing the Home-Brackman (HB) scale at 6-month post-operative follow-up visits. This was represented as a binary final result variable, post-operative preserved facial nerve operate (HB grade I) vs. post-operative facial nerve dysfunction (HB grades II–VI). Unbiased variables included patient-, tumor- and surgery-related traits, as described under.
Measurements of tumor dimensions had been made relative to the porus acusticus and posterior petrous bone (Supplementary Fig. 1)9. These dimensions had been chosen because of their relationships to surgical corridors and consistent with the aim of reproducibility in replicative efforts. Such measurements have additionally been proven to correlate nicely with volumetric analyses10. Measurements had been made by two raters, and settlement was assessed by intraclass correlation coefficient (ICC) accounting for 2-way random results11.
Normality of steady variables was assessed utilizing D’Agostino-Pearson’s check12, discovering that measurement C and case size had been usually distributed, and thus had been in contrast between HB I and HB II–VI teams with unbiased samples t-test. In distinction, age, BMI, measurements A, B, and D weren’t usually distributed and thus statistical significance of comparisons between facial nerve final result teams was assessed utilizing a Mann–Whitney U check. Categorical variables (intercourse, laterality, tumor measurement represented as a binary measurement of ≥ 2.5 cm vs. < 2.5 cm best tumor dimension, and presence/absence of residual tumor) had been evaluated for associations to the result utilizing Chi-squared assessments. All statistical assessments had been evaluated at a significance degree of alpha = 0.05.
Machine studying classifier choice and coaching
Research of machine studying proceed via sure regimented levels generally known as the machine studying lifecycle13. Though variations might exist based mostly on the precise examine and objectives, generally, the lifecycle begins with information assortment and pre-processing earlier than continuing via gathering of baseline descriptive statistical evaluation (described above), classifier choice, mannequin coaching, hyperparameter tuning, mannequin testing and finally deployment with the next assortment of further coaching examples for validation throughout deployment re-starting the cycle at information assortment (Supplementary Fig. 2).
To information classifier choice, the info had been first visualized by class distribution on every characteristic axis utilizing a pairplot (Supplementary Fig. 3). This demonstrated two essential traits of the info: the category imbalance was possible important sufficient to affect classifier efficiency and the info weren’t linearly separable on any two-dimensional characteristic axis airplane. Given the comparatively small measurement of the dataset, we utilized the artificial minority oversampling approach (SMOTE)14 to beat class imbalance: this offered new coaching examples that may be helpful in classifier coaching whereas equalizing the category distribution. Mannequin coaching then proceeded with collection of a classifier that was appropriate for the classification process whereas taking into consideration the restraints of the info. As a result of the info weren’t linearly separable, we chosen non-linear classifiers, together with the random forest, radial foundation operate (RBF) kernel assist vector machine (SVM), and synthetic neural community13 (Supplementary Fig. 2). Amongst these, the random forest classifier was chosen for additional growth because of its superior accuracy in performing the classification process on the coaching information. The info had been cut up for mannequin coaching (90%) and subsequent testing (10%). Whereas mannequin tuning was tried through hyperparameter optimization, the preliminary random forest mannequin with hyperparameters based mostly on the authors’ prior expertise with comparable classification duties and affected person datasets demonstrated the best accuracy.
The validation dataset (n = 32 sufferers) consisted solely of sufferers who underwent surgical procedure on the College of Pennsylvania within the ultimate yr of the examine, as this group had facial nerve outcomes assessed after preliminary algorithm growth and thus weren’t included within the preliminary coaching and testing information units. The identical affected person, tumor and surgical procedure traits had been collected for the 32 validation sufferers and the random forest algorithm was utilized to make predictions about which sufferers would have full facial nerve preservation vs. those that would have any facial nerve dysfunction. Predicted outcomes had been recorded and in comparison with precise 6-month facial nerve outcomes for this group of sufferers.
Outcomes
Two-hundred and forty-two consecutive sufferers had been recognized who underwent microsurgical resection of VS over the required time interval. Of those, 206 (85%) had preserved facial nerve operate (HB I), and 36 (15%) had any facial nerve dysfunction (HB II–VI). Abstract statistics and assessments of affiliation for underlying variations in patient-, tumor-, and surgery-specific traits between final result teams are proven in Desk 1. Among the many components evaluated, none demonstrated a statistically important distinction between the HB I and HB II–VI teams when evaluated on the idea of linear comparisons of measures of centrality (i.e., means and medians). The ICC for tumor measurements was between 90 and 99% for all measurements (Supplementary Fig. 1, Supplementary Desk 1).
When visualized in two dimensions, our information weren’t discovered to have linearly separable hyperplanes alongside any of the acquired characteristic axes (Supplementary Fig. 3). As such, non-linear supervised machine studying classifiers had been examined as described in Strategies (see additionally Supplementary Fig. 2). The random forest classifier carried out nicely with an accuracy of 90.5%15. Given the aim of making use of the classifier as a scientific device, sensitivity and specificity had been assessed on the check information, and had been discovered to be 90% and 90%, respectively. The receiver-operating attribute (ROC) curve is proven in Fig. 1A. A random sampling from a Gaussian distribution was generated as a random variable and used as a baseline to additional consider which options had been related within the random forest predictions: the ensuing characteristic importances had been computed and plotted (Fig. 1B). Relative to this baseline, the random forest classifier indicated a comparatively higher significance of BMI, case size, age, and measurement B, representing the extent of brainstem compression, in facial nerve operate prognostication. When examined on the validation information set, the mannequin demonstrated 84% accuracy in predicting facial nerve operate at 6 months post-operatively.
Dialogue
Facial nerve harm is a morbid complication of remedy for VS, with downstream results starting from social stigmata, affected person melancholy and lowered high quality of life16,17, to corneal abrasions and ulcers from incomplete eye closure and lack of corneal sensation18. Aside from tumor measurement, comparatively little is known about components which will affect facial nerve outcomes in microsurgery for VS. The scientific influence of facial nerve harm and significance of facial nerve preservation is highlighted by the in depth literature exploring predictors of facial nerve harm19,20,21,22,23. We leveraged our multi-institutional expertise at two facilities with excessive volumes of VS sufferers and utilized machine studying methods to establish novel predictors of facial nerve harm in sufferers handled with microsurgery.
Machine studying applied sciences have not too long ago undergone a resurgence alongside the event of computational instruments for dealing with and storing the massive quantities of knowledge required for his or her significant and broad scale utilization13,24. The popularity that such instruments can be utilized to glean novel tendencies from information that aren’t readily obvious from frequent descriptive statistical approaches makes their software throughout the scientific area a helpful and ongoing endeavor25. Such a phenomenon may be seen within the current examine the place assessments of affiliation, evaluating measures of centrality between final result teams, didn’t establish any components which considerably differed between sufferers with and with out preserved facial operate. In distinction, random forest characteristic significance evaluation discerned 4 options—BMI, case size, age and the tumor dimension representing progress in direction of the brainstem (measurement B)—as being related in predicting 6-month facial nerve standing. Whereas additional research have to be carried out to completely characterize the mechanistic position of those components in facial nerve final result, this demonstrates the utility of making use of novel information science methods to uncover non-linear interactions between variables which can have real-world, scientific relevance.
Tumor dimensions
As beforehand famous, tumor measurements utilized in our examine had been chosen because of their relationships to surgical corridors, in addition to having been proven to correlate nicely with tumor measurement by volumetric evaluation in earlier literature10. We discovered excessive ICC for all measurements, which was akin to different studies within the literature on comparable VS measurement duties26,27. Though traditionally, an general bigger tumor measurement has been demonstrated to portend worse facial nerve operate after microsurgical resection19,20,28,29,30, outcomes of the current examine recognized the tumor dimension representing progress throughout the cerebellopontine angle between the mid-axis of the tumor and the brainstem as most predictive of facial nerve final result. Our findings are per prior literature, whereas offering additional perception into attainable mechanisms by which tumor measurement might affect facial nerve harm. A comparatively bigger tumor dimension throughout the cerebellopontine angle, between the brainstem and porus acusticus is postulated to lead to extra thinning and splaying of the facial nerve. This causes direct mechanical harm and makes the facial nerve harder to tell apart from tumor capsule and surrounding adherent arachnoid, putting the facial nerve at higher danger of iatrogenic harm31. Thus, our examine builds on prior literature reporting higher tumor measurement as a predictor of facial nerve harm following vestibular schwannoma microsurgery, by suggesting that the tumor dimension representing progress throughout the cerebellopontine angle from the mid-axis of the tumor in direction of the brainstem has the best implication on facial nerve final result. We didn’t establish any distinction between our facial nerve preservation and facial nerve dysfunction teams when evaluating this dimension. It’s price noting that we noticed a comparatively greater fee of Koos grade III and IV tumors in comparison with different revealed sequence, suggesting that this sequence could also be skewed in direction of bigger tumors general. This will partially clarify our incapability to decipher a distinction between facial nerve preservation and facial nerve harm teams based mostly on tumor measurement. We anticipate that future research together with bigger cohorts of sufferers may seize a relationship between facial nerve susceptibility to harm as this tumor dimension will increase.
Age
Older affected person age has been beforehand proven to be predictive of facial nerve dysfunction, just like our personal findings20,29, although this stays controversial. Whereas some research have discovered no important relationship between post-operative facial nerve operate and age32, our examine and others have recognized a development in direction of rising age influencing unfavorable facial nerve outcomes following vestibular schwannoma microsurgery33. Others reporting on this discovering have hypothesized on the affect of frailty, burden of comorbidities, decreased neurologic reserve leading to lowered facial nerve rehabilitation potential33, and the confounding affect of age itself on facial nerve grading on condition that pores and skin laxity and thinning might contribute to worse grading and/or worsened manifestations of facial nerve paralysis in aged sufferers34. We additional hypothesize that the idea of this relationship is perhaps much less favorable tissue dissection planes in sufferers of superior age, putting older sufferers at higher danger of iatrogenic facial nerve harm. Though additional detailed evaluation of the position of age in facial nerve final result on sufferers present process vestibular schwannoma microsurgery is past the scope of the present examine, additional examine will surely be helpful to verify and higher characterize the character of this relationship. Our examine additional demonstrated further distinctive options predictive of facial nerve outcomes which haven’t been beforehand recognized. Our hypotheses concerning the position of BMI and case size are mentioned additional under.
BMI
Curiously, our mannequin recognized BMI and operative case size as being extremely predictive of facial nerve final result at 6 months post-operatively. To the perfect of our information, these associations haven’t been clearly delineated in earlier research. One examine examined facial nerve harm within the context of post-operative problems and the necessity for readmission or re-operation, discovering no important affiliation to BMI35. Nonetheless, because the authors observe, facial nerve harm usually happens with out the requirement for reoperation and readmission, thus is probably going underrepresented of their evaluation. One other examine evaluated the affect of BMI on imply HB rating pre-operatively (1.1 non-obese vs. 1.0 overweight, p = 0.16) and post-operatively (1.9 non-obese vs. 1.7 overweight, p = 0.32) discovering no distinction between overweight and non-obese teams36. Nonetheless, the timing of facial nerve operate evaluation is just not clearly specified on this examine and when facial operate is modelled as a categorical variable (moderately than steady, summarized with imply HB scores), overweight sufferers had been extra possible than non-obese sufferers to have HB scores equal to or higher than III (9.2% non-obese vs. 17.7% overweight). The noticed affiliation between BMI and facial nerve dysfunction in our examine could also be seen as hypothesis-generating, and ought to be explored in future research. It’s attainable that troublesome surgical ergonomics in high-BMI sufferers make tumor dissection off of the facial nerve harder, putting sufferers at greater danger of dysfunction37,38,39. For instance, in greater BMI sufferers, comparatively greater mass of the neck and shoulder might additional slim an already small operative working hall, which along with requiring much less ergonomic positioning for tumor entry, limits the dissection vectors and angles, and reduces vary of movement and visibility. The elevated utilization of endoscopes40 and exoscopes41 in lateral cranium base surgical procedure might ultimately mitigate a few of these constraints.
Case size
Operative length is recognized as a key issue related to facial nerve final result in microsurgical resection of vestibular schwannomas within the current examine—to our information, that is the primary such description of this affiliation, nonetheless, that is per earlier research wherein extended operative length has been proven to be related to a better fee of problems42. Our noticed affiliation of elevated operative size being related to a better chance of facial nerve dysfunction could also be reflective in a part of the recognized affiliation between tumor measurement and facial nerve outcomes, because of bigger tumors having longer common operative durations. Nonetheless, on condition that bigger general tumor measurement and particular person tumor measurements in three dimensions (parallel to the posterior petrous bone, between central axis of tumor and porus acusticus, and from porus acusticus to distalmost extent of tumor progress throughout the IAC) weren’t discovered to be predictive of facial nerve dysfunction, different components which can improve case size ought to be thought-about and investigated in future research because the underlying mechanism of this affiliation. Components reminiscent of tumor hypervascularity43, adherence to the facial nerve perineurium, and the path of facial nerve displacement could also be mirrored amongst distinction in operative size throughout sufferers, and thus contribute to the noticed differential danger of facial nerve dysfunction because it pertains to case size20. These components might function a surrogate for dissection complexity. Lastly, you will need to acknowledge that this algorithm, as any machine studying/synthetic intelligence device, is proscribed by the inputs. As such, there could also be different confounding variables that affect facial nerve harm danger which weren’t captured in our information or evaluation. Additional examine will likely be important to raised perceive the myriad components which can affect the position of case size on facial nerve final result in vestibular schwannoma microsurgery.
A significant energy of this examine is the inclusion of affected person cohorts from three hospitals throughout two well being methods, rising the generalizability of the ensuing mannequin. The mannequin demonstrates an anticipated efficiency decay from 90.5 to 84% when assessed on unseen information from one of many included establishments. This degree of efficiency decay each demonstrates the low chance of overfitting of this mannequin and the relative reliability of the mannequin in the true world (scientific) context. Whereas the present mannequin demonstrates good accuracy whereas avoiding overfitting, we acknowledge that efficiency will proceed to enhance within the deployment part as additional information is collected at exterior websites and thru future potential validation with affected person information from the taking part establishments (Supplementary Fig. 2). Whereas we respect the large good thing about multi-center information assortment to boost reproducibility, generalizability and scientific translation of our algorithm, we additionally acknowledge that as we improve the variety of taking part facilities and broaden to incorporate establishments outdoors of our area, hospital-related components (setting, degree of care, gear, and so on.) and surgeon-related components (affected person choice, most popular surgical strategy, years of expertise, and so on.), will have to be thought-about and evaluated on this stage of deployment44.
A limitation of the current examine is an general small proportion of sufferers with facial nerve dysfunction, which possible restricted the statistical significance of associations which can have scientific relevance, in addition to our means to additional stratify sufferers into completely different grades of facial operate (i.e. HB I–VI). As vestibular schwannoma is a comparatively uncommon illness entity, increasing our database with every at the moment taking part establishment will happen at a fee of roughly 30–60 sufferers per yr, thus rising the time to construct a dataset sturdy sufficient to meaningfully enhance the mannequin metrics and generalizability. Nonetheless, we intention to beat this limitation via dissemination of our outcomes and the present iteration of the algorithm—we intention to broaden this work to incorporate further intuitions each nationally and internationally with the objectives of enhancing statistical energy, and additional rising the generalizability of this work. As further validation is carried out, we anticipate that the machine studying lifecycle will re-start, together with additional iterations of mannequin analysis and tuning to additional enhance efficiency.
As beforehand famous, the present iteration of this algorithm was developed based mostly on guide tumor measurements which were proven to have sturdy reproducibility and correlation with volumetric evaluation all through the vestibular schwannoma literature. Nonetheless, accelerated deployment might be expedited via automated tumor segmentation—a number of such promising instruments have not too long ago been developed for vestibular schwannoma, nonetheless, in all circumstances the authors acknowledge that these would require additional validation earlier than implementation45,46,47,48. This strategy has proven important promise in different medical contexts, significantly in creating methods for automating chest X-ray overview through the COVID-19 pandemic49,50, and within the identification of regarding vs. benign gastrointestinal polyps51,52. Lastly, as information science methods are more and more utilized in medication, no dialogue of their implementation on this context is full with out contemplating the safety of affected person privateness and confidentiality. The algorithm we current right here is run regionally and utterly offline. Nonetheless, cloud-based automation gives a number of benefits that have to be weighed towards the potential for information leakage—methods for obviating safety considerations whereas sustaining the flexibleness, reliability, and accelerated deployment afforded by these instruments are underneath growth. A full dialogue of such strategies is past the scope of this paper, however may be additional explored in current works by Mei et al.53 and Wu et al.54, amongst others.
It’s our aim that this algorithm will finally be utilized as a clinically helpful device for stratifying a person affected person’s danger of facial nerve harm, aiding in pre-operative counseling about remedy strategy (watchful ready vs. radiosurgery vs. microsurgical resection) and timing. Importantly, the mannequin was evaluated through accuracy, sensitivity and specificity given the frequent utilization of those as metrics of check efficiency within the scientific setting. On this particular context, we interpret the 90% accuracy to be wonderful in comparison with the 85% accuracy which has been referenced as a benchmark of acceptable efficiency15—we additional anticipate improved accuracy and generalizability efficiency (much less efficiency decay), with the addition of validation examples throughout deployment. As well as, the sensitivity and specificity of 90% and 90% characterize that the mannequin performs equally nicely at predicting which sufferers are more likely to have full facial nerve preservation because it does at predicting which sufferers are more likely to have facial nerve dysfunction. We anticipate that additional validation via collaboration with further facilities which deal with excessive volumes of vestibular schwannomas will proceed to enhance the mannequin’s efficiency.
Recognizing that clinicians and sufferers with little to no pc programming background might discover it cumbersome to implement the algorithm, we plan to develop a graphical person interface to facilitate ease of use in each exploratory and scientific settings. This idea has been utilized in different areas of drugs to facilitate a user-friendly implementation of synthetic intelligence within the scientific setting55,56.
Future instructions
Historically, tumor measurement has been the one most essential think about counseling sufferers concerning their danger of facial nerve harm. Importantly, our findings recommend that further patient-, tumor- and surgery-related components may affect the chance of facial nerve harm in vestibular schwannoma microsurgery. The discovering that the tumor dimension representing the mid-axis of the tumor to the brainstem is essential in deciphering which sufferers are more likely to expertise facial nerve harm builds on present literature which has discovered tumor measurement to be a important determinant of facial nerve final result by providing extra granularity to the outline of the potential position of tumor measurement. As well as, our discovering that elevated age is of relative significance in predicting facial nerve final result provides to present literature discovering the identical. Lastly, the findings that elevated BMI and longer case size are of relative significance in predicting the chance of facial nerve dysfunction following vestibular schwannoma microsurgery are novel to this examine and hypothesis-generating. For the entire described components, future validation in unbiased cohorts are worthwhile endeavors. As well as, additional exploration of variables not represented on this examine, however which could affect facial nerve final result in vestibular schwannoma microsurgery might proceed to construct on the findings introduced right here in direction of enhancing affected person outcomes.
Conclusions
Right here, we’ve described the event of a multi-institutionally derived machine studying algorithm to foretell the chance of facial nerve harm following microsurgical resection of VS. Our mannequin demonstrated a excessive diploma of accuracy, and was in a position to establish novel predictors of facial nerve dysfunction following microsurgical resection of VS. The mannequin will likely be additional developed as a scientific device for predictions of facial nerve final result. With the inclusion of further nationwide and worldwide establishments to enhance generalizability, our final aim is to make the most of this device for counseling sufferers about surgical danger, and help in surgical decision-making. Extra broadly, whereas additional analysis is important to completely perceive the mechanistic implications of the options recognized, this evaluation has demonstrated the utility of machine studying in figuring out clinically related components which can in any other case evade elucidation through linear statistical strategies, reminiscent of comparisons of measures of centrality.
Information availability
In accordance with the College of Pennsylvania Institutional Overview Board necessities for examine exemption, the analysis information supporting this undertaking will not be publicly shared. De-identified information could also be shared upon affordable request following execution of a Information Utilization Settlement. Please attain out to corresponding writer, SMHA, with any inquiries.
References
-
Acoustic Neuroma. NORD. https://rarediseases.org/rare-diseases/acoustic-neuroma/ (accessed 29 Nov 2021).
-
Vestibular Schwannoma (Acoustic Neuroma) and Neurofibromatosis. NIDCD. https://www.nidcd.nih.gov/well being/vestibular-schwannoma-acoustic-neuroma-and-neurofibromatosis (accessed 29 Nov 2021).
-
Mohammadi, A. & Jufas, N. Sudden loss of life because of vestibular schwannoma: Warning in emergent administration. Otol. Neurotol. 37(5), 564–567. https://doi.org/10.1097/mao.0000000000001004 (2016).
Google Scholar
-
Lin, E. P. & Crane, B. T. The administration and imaging of vestibular schwannomas. Am. J. Neuroradiol. 38(11), 2034. https://doi.org/10.3174/ajnr.A5213 (2017).
Google Scholar
-
Ferroli, P., Bosio, L. & Broggi, M. Facial nerve sparing surgical procedure for giant vestibular schwannomas. Acta Neurochir. (Wien). 159(7), 1213–1218. https://doi.org/10.1007/s00701-017-3216-y (2017).
Google Scholar
-
Roland, J. T. Jr. et al. Cranial nerve preservation in surgical procedure for giant acoustic neuromas. Cranium Base. 14(2), 85–91. https://doi.org/10.1055/s-2004-828699 (2004).
Google Scholar
-
Profant, O. et al. Resolution making on vestibular schwannoma remedy: Predictions based mostly on machine-learning evaluation. Sci. Rep. 11(1), 18376. https://doi.org/10.1038/s41598-021-97819-x (2021).
Google Scholar
-
Cha, D., Shin, S. H., Kim, S. H., Choi, J. Y. & Moon, I. S. Machine studying strategy for prediction of listening to preservation in vestibular schwannoma surgical procedure. Sci. Rep. 10(1), 7136. https://doi.org/10.1038/s41598-020-64175-1 (2020).
Google Scholar
-
Slattery, W. H. third., Fisher, L. M., Yoon, G., Sorensen, G. & Lev, M. Magnetic resonance imaging scanner reliability for measuring modifications in vestibular schwannoma measurement. Otol. Neurotol. 24(4), 666–670. https://doi.org/10.1097/00129492-200307000-00022 (2003) (dialogue 670–1).
Google Scholar
-
Choi, Y. et al. Most diameter versus volumetric evaluation for the response analysis of vestibular schwannomas receiving stereotactic radiotherapy. Radiat. Oncol. J. 36(2), 114–121. https://doi.org/10.3857/roj.2018.00031 (2018).
Google Scholar
-
Koo, T. Okay. & Li, M. Y. A tenet of choosing and reporting intraclass correlation coefficients for reliability analysis. J. Chiropr. Med. 15(2), 155–163. https://doi.org/10.1016/j.jcm.2016.02.012 (2016).
Google Scholar
-
D’Agostino, R. & Pearson, E. S. Assessments for departure from normality. Biometrika. 60(3), 613–622. https://doi.org/10.2307/2335012 (1973).
Google Scholar
-
Sebastian Raschka, V. M. Python Machine Studying: Machine Studying and Deep Studying with Python, scikit-learn and TensorFlow 2. third edn (Packt Publishing Ltd., 2019).
-
Chawla, N. V., Bowyer, Okay. W., Corridor, L. O. & Kegelmeyer, W. P. SMOTE: Artificial minority over-sampling approach. J. Artif. Int. Res. 16(1), 321–357 (2002).
-
Wilson, R. C., Shenhav, A., Straccia, M. & Cohen, J. D. The eighty 5 p.c rule for optimum studying. Nat. Commun. 10(1), 4646. https://doi.org/10.1038/s41467-019-12552-4 (2019).
Google Scholar
-
Dobel, C., Miltner, W. H., Witte, O. W., Volk, G. F. & Guntinas-Lichius, O. Emotional influence of facial palsy. Laryngorhinootologie. 92(1), 9–23. https://doi.org/10.1055/s-0032-1327624 (2013) (Emotionale Auswirkungen einer Fazialisparese).
Google Scholar
-
Saadi, R., Shokri, T., Schaefer, E., Hollenbeak, C. & Lighthall, J. G. Melancholy charges after facial paralysis. Ann. Plast. Surg. 83(2), 190–194. https://doi.org/10.1097/sap.0000000000001908 (2019).
Google Scholar
-
Joseph, S. S. et al. Analysis of sufferers with facial palsy and ophthalmic sequelae: A 23-year retrospective overview. Ophthalmic Epidemiol. 24(5), 341–345. https://doi.org/10.1080/09286586.2017.1294186 (2017).
Google Scholar
-
Ren, Y., MacDonald, B. V., Tawfik, Okay. O., Schwartz, M. S. & Friedman, R. A. Medical predictors of facial nerve outcomes after surgical resection of vestibular schwannoma. Otolaryngol. Head Neck Surg. 164(5), 1085–1093. https://doi.org/10.1177/0194599820961389 (2021).
Google Scholar
-
Solar, Y., Yang, J., Li, T., Gao, Okay. & Tong, X. Nomogram for predicting facial nerve outcomes after surgical resection of vestibular schwannoma. Entrance. Neurol. 12, 817071. https://doi.org/10.3389/fneur.2021.817071 (2021).
Google Scholar
-
Tawfik, Okay. O., Alexander, T. H., Saliba, J., Mastrodimos, B. & Cueva, R. A. Predicting long-term facial nerve outcomes after resection of vestibular schwannoma. Otol. Neurotol. 41(10), e1328–e1332. https://doi.org/10.1097/mao.0000000000002883 (2020).
Google Scholar
-
Troude, L. et al. Predictive components of early postoperative and long-term facial nerve operate after giant vestibular schwannoma surgical procedure. World Neurosurg. 127, e599–e608. https://doi.org/10.1016/j.wneu.2019.03.218 (2019).
Google Scholar
-
Axon, P. R. & Ramsden, R. T. Intraoperative electromyography for predicting facial operate in vestibular schwannoma surgical procedure. Laryngoscope. 109(6), 922–926. https://doi.org/10.1097/00005537-199906000-00015 (1999).
Google Scholar
-
Rajkomar, A., Dean, J. & Kohane, I. Machine studying in medication. N. Engl. J. Med. 380(14), 1347–1358. https://doi.org/10.1056/NEJMra1814259 (2019).
Google Scholar
-
Karargyris, A. et al. Federated benchmarking of medical synthetic intelligence with MedPerf. Nat. Mach. Intell. 5(7), 799–810. https://doi.org/10.1038/s42256-023-00652-2 (2023).
Google Scholar
-
Erickson, N. J. et al. Koos classification of vestibular schwannomas: A reliability examine. Neurosurgery. 85(3), 409–414. https://doi.org/10.1093/neuros/nyy409 (2019).
Google Scholar
-
Lawson McLean, A. C., McLean, A. L. & Rosahl, S. Okay. Evaluating vestibular schwannoma measurement and quantity on magnetic resonance imaging: An inter- and intra-rater settlement examine. Clin. Neurol. Neurosurg. 145, 68–73. https://doi.org/10.1016/j.clineuro.2016.04.010 (2016).
Google Scholar
-
Fenton, J. E., Chin, R. Y., Fagan, P. A., Sterkers, O. & Sterkers, J. M. Predictive components of long-term facial nerve operate after vestibular schwannoma surgical procedure. Otol. Neurotol. 23(3), 388–392. https://doi.org/10.1097/00129492-200205000-00027 (2002).
Google Scholar
-
Rivas, A. et al. A mannequin for early prediction of facial nerve restoration after vestibular schwannoma surgical procedure. Otol. Neurotol. 32(5), 826–833. https://doi.org/10.1097/MAO.0b013e31821b0afd (2011).
Google Scholar
-
Moffat, D. A., Parker, R. A., Hardy, D. G. & Macfarlane, R. Components affecting ultimate facial nerve final result following vestibular schwannoma surgical procedure. J. Laryngol. Otol. 128(5), 406–415. https://doi.org/10.1017/s0022215114000541 (2014).
Google Scholar
-
Raslan, A. M., Liu, J. Okay., McMenomey, S. O. & Delashaw, J. B. Jr. Staged resection of enormous vestibular schwannomas. J. Neurosurg. 116(5), 1126–1133. https://doi.org/10.3171/2012.1.Jns111402 (2012).
Google Scholar
-
Bloch, O. et al. Components related to preservation of facial nerve operate after surgical resection of vestibular schwannoma. J. Neuro-Oncol. 102(2), 281–286. https://doi.org/10.1007/s11060-010-0315-5 (2011).
Google Scholar
-
Helal, A. et al. Differential influence of superior age on scientific outcomes after vestibular schwannoma resection within the very aged: Cohort examine. Oper. Neurosurg. (Hagerstown). 21(3), 104–110. https://doi.org/10.1093/ons/opab170 (2021).
Google Scholar
-
Macielak, R. J. et al. The impact of age on facial nerve restoration after vestibular schwannoma resection. Otol. Neurotol. 44(7), 725–729. https://doi.org/10.1097/mao.0000000000003937 (2023).
Google Scholar
-
Murphy, M. E. et al. Morbid weight problems will increase danger of morbidity and reoperation in resection of benign cranial nerve neoplasms. Clin. Neurol. Neurosurg. 148, 105–109. https://doi.org/10.1016/j.clineuro.2016.06.020 (2016).
Google Scholar
-
Lipschitz, N. et al. Weight problems is just not related to postoperative problems after vestibular schwannoma surgical procedure in a big single establishment sequence. Otol. Neurotol. 40(10), 1373–1377. https://doi.org/10.1097/mao.0000000000002397 (2019).
Google Scholar
-
Moss, E. L. et al. Affect of weight problems on surgeon ergonomics in robotic and straight-stick laparoscopic surgical procedure. J. Minim. Invasive Gynecol. 27(5), 1063–1069. https://doi.org/10.1016/j.jmig.2019.07.009 (2020).
Google Scholar
-
Sers, R., Forrester, S., Zecca, M., Ward, S. & Moss, E. The influence of affected person physique mass index on surgeon posture throughout simulated laparoscopy. medRxiv. https://doi.org/10.1101/2020.11.24.20237123 (2020).
Google Scholar
-
Sers, R., Forrester, S., Zecca, M., Ward, S. & Moss, E. Goal evaluation of surgeon kinematics throughout simulated laparoscopic surgical procedure: A preliminary analysis of the impact of excessive physique mass index fashions. Int. J. Comput. Help. Radiol. Surg. 17(1), 75–83. https://doi.org/10.1007/s11548-021-02455-5 (2022).
Google Scholar
-
Setty, P. et al. Endoscopic resection of vestibular schwannomas. J. Neurol. Surg. B Cranium Base. 76(3), 230–238. https://doi.org/10.1055/s-0034-1543974 (2015).
Google Scholar
-
Veldeman, M. et al. Three-dimensional exoscopic versus microscopic resection of vestibular schwannomas: A comparative sequence. Oper. Neurosurg. (Hagerstown). 24(5), 507–513. https://doi.org/10.1227/ons.0000000000000602 (2023).
Google Scholar
-
Cheng, H. et al. Extended operative length is related to problems: A scientific overview and meta-analysis. J. Surg. Res. 229, 134–144. https://doi.org/10.1016/j.jss.2018.03.022 (2018).
Google Scholar
-
Teranishi, Y., Kohno, M., Sora, S., Sato, H. & Nagata, O. Hypervascular vestibular schwannomas: Medical traits, angiographical classification, and surgical concerns. Oper. Neurosurg. 15(3), 251–261 (2018).
Google Scholar
-
Jo, D. The interpretation bias and lure of multicenter scientific analysis. Korean J. Ache. 33(3), 199–200. https://doi.org/10.3344/kjp.2020.33.3.199 (2020).
Google Scholar
-
Neve, O. M. et al. Totally automated 3D vestibular schwannoma segmentation with and with out gadolinium-based distinction materials: A multicenter, multivendor examine. Radiol. Artif. Intell. 4(4), e210300. https://doi.org/10.1148/ryai.210300 (2022).
Google Scholar
-
Shapey, J. et al. Information from: Segmentation of vestibular schwannoma from magnetic resonance imaging: An open annotated dataset and baseline algorithm. Sci. Information https://doi.org/10.7937/TCIA.9YTJ-5Q73 (2021).
Google Scholar
-
Shapey, J. et al. A synthetic intelligence framework for automated segmentation and volumetry of vestibular schwannomas from contrast-enhanced T1-weighted and high-resolution T2-weighted MRI. J. Neurosurg. JNS. 134(1), 171–179. https://doi.org/10.3171/2019.9.JNS191949 (2021).
Google Scholar
-
Wang, H., Qu, T., Bernstein, Okay., Barbee, D. & Kondziolka, D. Automated segmentation of vestibular schwannomas from T1-weighted MRI with a deep neural community. Radiat. Oncol. 18(1), 78. https://doi.org/10.1186/s13014-023-02263-y (2023).
Google Scholar
-
Su, H. et al. Multilevel threshold picture segmentation for COVID-19 chest radiography: A framework utilizing horizontal and vertical multiverse optimization. Comput. Biol. Med. 146, 105618. https://doi.org/10.1016/j.compbiomed.2022.105618 (2022).
Google Scholar
-
Qi, A. et al. Directional mutation and crossover boosted ant colony optimization with software to COVID-19 X-ray picture segmentation. Comput. Biol. Med. 148, 105810. https://doi.org/10.1016/j.compbiomed.2022.105810 (2022).
Google Scholar
-
Hu, Okay. et al. Colorectal polyp area extraction utilizing saliency detection community with neutrosophic enhancement. Comput. Biol. Med. 147, 105760. https://doi.org/10.1016/j.compbiomed.2022.105760 (2022).
Google Scholar
-
Jiang, X. et al. BiFTransNet: A unified and simultaneous segmentation community for gastrointestinal pictures of CT & MRI. Comput. Biol. Med. 165, 107326. https://doi.org/10.1016/j.compbiomed.2023.107326 (2023).
Google Scholar
-
Mei, Z. et al. Safe multi-dimensional information retrieval with entry management and vary question within the cloud. Inf. Syst. 122, 102343. https://doi.org/10.1016/j.is.2024.102343 (2024).
Google Scholar
-
Wu, Z. et al. An efficient methodology for the safety of person well being matter privateness for well being data providers. World Broad Net. 26(6), 3837–3859. https://doi.org/10.1007/s11280-023-01208-5 (2023).
Google Scholar
-
She, Y. et al. Improvement and validation of a deep studying mannequin for non-small cell lung most cancers survival. JAMA Netw. Open. 3(6), e205842. https://doi.org/10.1001/jamanetworkopen.2020.5842 (2020).
Google Scholar
-
Ye, R. Z. et al. DeepImageTranslator: A free, user-friendly graphical interface for picture translation utilizing deep-learning and its purposes in 3D CT picture evaluation. SLAS Technol. 27(1), 76–84. https://doi.org/10.1016/j.slast.2021.10.014 (2022).
Google Scholar
Funding
This work was supported by a grant from the Nationwide Institutes of Well being, USA (SMHA, NIH grant T32NS091006-07) and a Dean’s Grasp’s Scholarship from the College of Pennsylvania Division of Bioengineering (SMHA).
Creator data
Authors and Affiliations
Contributions
S.M.H.A.: conceived and designed the examine, collected information, contributed information or evaluation instruments, carried out information evaluation, wrote the manuscript. R.B.: collected information, contributed information or evaluation instruments, reviewed the manuscript. A.E.Q.: contributed information or evaluation instruments, reviewed the manuscript. H.A.: collected information, contributed information or evaluation instruments, reviewed the manuscript. E.M.S.: contributed information or evaluation instruments, reviewed the manuscript. D.C.: contributed information or evaluation instruments, reviewed the manuscript. T.H.: reviewed the manuscript. J.B.: reviewed the manuscript. M.J.R.: reviewed the manuscript. D.C.B.: reviewed the manuscript. C.J.: reviewed the manuscript. G.Z.: contributed information or evaluation instruments, reviewed the manuscript. P.G.: contributed information or evaluation instruments, reviewed the manuscript. S.E.B.: conceived and designed the examine, reviewed the manuscript. Y.E.C.: contributed information or evaluation instruments, reviewed the manuscript. J.Y.Okay.L.: contributed information or evaluation instruments, reviewed the manuscript.
Corresponding writer
Ethics declarations
Competing pursuits
The authors declare no competing pursuits.
Extra data
Writer’s observe
Springer Nature stays impartial with regard to jurisdictional claims in revealed maps and institutional affiliations.
Supplementary Data
Supplementary Data.
Rights and permissions
Open Entry This text is licensed underneath a Inventive Commons Attribution 4.0 Worldwide License, which allows use, sharing, adaptation, distribution and copy in any medium or format, so long as you give applicable credit score to the unique writer(s) and the supply, present a hyperlink to the Inventive Commons licence, and point out if modifications had been made. The photographs or different third social gathering materials on this article are included within the article’s Inventive Commons licence, except indicated in any other case in a credit score line to the fabric. If materials is just not included within the article’s Inventive Commons licence and your supposed use is just not permitted by statutory regulation or exceeds the permitted use, you’ll need to acquire permission immediately from the copyright holder. To view a duplicate of this licence, go to http://creativecommons.org/licenses/by/4.0/.
Reprints and permissions
About this text
Cite this text
Heman-Ackah, S.M., Blue, R., Quimby, A.E. et al. A multi-institutional machine studying algorithm for prognosticating facial nerve harm following microsurgical resection of vestibular schwannoma.
Sci Rep 14, 12963 (2024). https://doi.org/10.1038/s41598-024-63161-1
-
Acquired: 18 November 2023
-
Accepted: 26 Might 2024
-
Revealed: 05 June 2024
-
DOI: https://doi.org/10.1038/s41598-024-63161-1
Feedback
By submitting a remark you conform to abide by our Phrases and Group Tips. In case you discover one thing abusive or that doesn’t adjust to our phrases or tips please flag it as inappropriate.
Adblock check (Why?)