Deep learning-based prediction of post-pancreaticoduodenectomy pancreatic fistula.

Postoperative pancreatic fistula is a life-threatening complication with an unmet need for accurate prediction. This study was aimed to develop preoperative artificial intelligence-based prediction models. Patients who underwent pancreaticoduodenectomy were enrolled and stratified into model development and validation sets by surgery between 2016 and 2017 or in 2018, respectively. Machine learning models based on clinical and body composition data, and deep learning models based on computed tomographic data, were developed, combined by ensemble voting, and final models were selected comparison with earlier model. Among the 1333 participants (training, n = 881; test, n = 452), postoperative pancreatic fistula occurred in 421 (47.8%) and 134 (31.8%) and clinically relevant postoperative pancreatic fistula occurred in 59 (6.7%) and 27 (6.0%) participants in the training and test datasets, respectively. In the test dataset, the area under the receiver operating curve [AUC (95% confidence interval)] of the selected preoperative model for predicting all and clinically relevant postoperative pancreatic fistula was 0.75 (0.71-0.80) and 0.68 (0.58-0.78). The ensemble model showed better predictive performance than the individual ML and DL models.

Preoperative data-based deep learning model for predicting postoperative survival in pancreatic cancer patients.

BACKGROUND: Pancreatic ductal adenocarcinoma (PDAC) has a poor prognosis even after curative resection. A deep learning-based stratification of postoperative survival in the preoperative setting may aid the treatment decisions for improving prognosis. This study was aimed to develop a deep learning model based on preoperative data for predicting postoperative survival. METHODS: The patients who underwent surgery for PDAC between January 2014 and May 2015. Clinical data-based machine learning models and computed tomography (CT) data-based deep learning models were developed separately, and ensemble learning was utilized to combine two models. The primary outcomes were the prediction of 2-year overall survival (OS) and 1-year recurrence-free survival (RFS). The model's performance was measured by area under the receiver operating curve (AUC) and was compared with that of American Joint Committee on Cancer (AJCC) 8th stage. RESULTS: The median OS and RFS were 23 and 10 months in training dataset (n = 229), and 22 and 11 months in test dataset (n = 53), respectively. The AUC of the ensemble model for predicting 2-year OS and 1-year RFS in the test dataset was 0.76 and 0.74, respectively. The performance of the ensemble model was comparable to that of the AJCC in predicting 2-year OS (AUC, 0.67; P = 0.35) and superior to the AJCC in predicting 1-year RFS (AUC, 0.54; P = 0.049). CONCLUSION: Our ensemble model based on routine preoperative variables showed good performance for predicting prognosis for PDAC patients after surgery.

Prediction of Life-Threatening Intracranial Hypertension During the Acute Phase of Traumatic Brain Injury Using Machine Learning.

Intracranial hypertension (IH) following acute phase traumatic brain injury (TBI) is associated with high mortality. Objective: This study proposes a novel parameter that may identify a potentially life-threatening IH (LTH) event and designs a machine learning model to predict LTH. Continuous recordings of intracranial pressure (ICP) and arterial blood pressure (ABP) from 273 TBI patients were used as the development dataset. The pressure-time dose (PTD) and pressure reactivity index (PRx) were calculated for each IH event, and an IH event with PRx > 0 and PTD > 5 was considered an LTH event. The association between the LTH parameters accumulated over five days and mortality was analyzed. A categorical boosting (CatBoost) model was employed to predict the occurrence of a future LTH event from the onset of IH using the ABP- and ICP-related parameters. Training and validation were performed on a total of 5,938 IH events. External performance evaluation was performed in 307 IH events included in the Cerebral Haemodynamic Autoregulatory Information System (CHARIS) database. The performance of the proposed model was evaluated through the area under the receiver operating characteristic curve (AUROC). The LTH parameters were able to distinguish between the deceased and surviving patients (AUROC > 0.7, p < 0.001). The CatBoost model predicted LTH with an AUROC = 0.7 on the external test dataset. This study demonstrated that the proposed LTH prediction model has a reasonable predictive capacity for mortality. The CatBoost model anticipates whether an IH event will develop into an LTH event. The findings of this study support the usefulness of ICP monitoring.

Novel index for predicting mortality during the first 24 hours after traumatic brain injury.

OBJECTIVE: Failure of cerebral autoregulation and subsequent hypoperfusion is common during the acute phase of traumatic brain injury (TBI). The cerebrovascular pressure-reactivity index (PRx) indirectly reflects cerebral autoregulation and has been used to derive optimal cerebral perfusion pressure (CPP). This study provides a method for the use of a combination of PRx, CPP, and intracranial pressure (ICP) to better evaluate the extent of cerebral hypoperfusion during the first 24 hours after TBI, allowing for a more accurate prediction of mortality risk. METHODS: Continuous ICP and arterial blood pressure (ABP) signals acquired from 295 TBI patients during the first 24 hours after admission were retrospectively analyzed. The CPP at the lowest PRx was determined as the optimal CPP (CPPopt). The duration of a severe hypoperfusion event (dHP) was defined as the cumulative time that the PRx was > 0.2 and the CPP was < 70 mm Hg with the addition of intracranial hypertension (ICP > 20 or > 22 mm Hg). The outcome was determined as 6-month mortality. RESULTS: The cumulative duration of PRx > 0.2 and CPP < 70 mm Hg exhibited a significant association with mortality (p < 0.001). When utilized with basic clinical information available during the first 24 hours after admission (i.e., Glasgow Coma Scale score, age, and mean ICP), a dHP > 25 minutes yielded a significant predictive capacity for mortality (p < 0.05, area under the curve [AUC] = 0.75). The parameter was particularly predictive of mortality for patients with a mean ICP > 20 or > 22 mm Hg (AUC = 0.81 and 0.87, respectively). CONCLUSIONS: A short duration (25 minutes) of severe hypoperfusion, evaluated as lowered CPP during worsened cerebrovascular reactivity during the 1st day after TBI, is highly indicative of mortality.

Cerebrospinal fluid dynamics at the lumbosacral level in patients with spinal stenosis: A pilot study.

Spinal stenosis is a common degenerative condition. However, how neurogenic claudication develops has not been clearly elucidated. Moreover, cerebrospinal fluid physiology at the lumbosacral level has not received adequate attention. This study was conducted to compare cerebrospinal fluid hydrodynamics at the lumbosacral spinal level between patients with spinal stenosis and healthy controls. Twelve subjects (four patients and eight healthy controls; 25-77 years old; seven males) underwent phase-contrast magnetic resonance imaging to quantify cerebrospinal fluid dynamics. The cerebrospinal fluid flow velocities were measured at the L2 and S1 levels. All subjects were evaluated at rest and after walking (to provoke neurogenic claudication in the patients). The caudal peak flow velocity in the sacral spine (-0.25 ± 0.28 cm/s) was attenuated compared to that in the lumbar spine (-0.93 ± 0.46 cm/s) in both patients and controls. The lumbar caudal peak flow velocity was slower in patients (-0.65 ± 0.22 cm/s) than controls (-1.07 ± 0.49 cm/s) and this difference became more pronounced after walking (-0.66 ± 0.37 cm/s in patients, -1.35 ± 0.52 cm/s in controls; p = 0.028). The sacral cerebrospinal fluid flow after walking was barely detectable in patients (caudal peak flow velocity: -0.09 ± 0.03 cm/s). Cerebrospinal fluid dynamics in the lumbosacral spine were more attenuated in patients with spinal stenosis than healthy controls. After walking, the patients experiencing claudication did not exhibit an increase in the cerebrospinal fluid flow rate as the controls did. Altered cerebrospinal fluid dynamics may partially explain the pathophysiology of spinal stenosis. © 2016 Orthopaedic Research Society. Published by Wiley Periodicals, Inc. J Orthop Res 35:104-112, 2017.

Finite Element Model for Hydrocephalus and Idiopathic Intracranial Hypertension.

Hydrocephalus and idiopathic intracranial hypertension (IIH) are neuropathies associated with disturbed cerebrospinal fluid dynamics. Several finite element (FE) brain models were suggested to simulate the pathological changes in hydrocephalus, but with overly simplified assumptions regarding the properties of the brain parenchyma. This study proposes a two-dimensional FE brain model, capable of simulating both hydrocephalus and IIH by incorporating poro-hyperelasticity of the brain and detailed structural information (i.e., sulci).

Morphological Feature Extraction From a Continuous Intracranial Pressure Pulse via a Peak Clustering Algorithm.

OBJECTIVE: An increase in intracranial pressure (ICP) is frequently observed in patients with severe traumatic brain injury (TBI). The information derived from the observation of temporal changes in the mean ICP is insufficient for assessment of the compensatory reserve of the injured brain. This assessment can be achieved via continuous morphological analysis of the pulse waveform of the ICP. METHODS: Continuous arterial blood pressure (ABP) and ICP recordings from 292 TBI patients were analyzed. The algorithm extracted morphological landmarks (peaks, troughs, and flats) from the ICP. Among the extracted peaks, P1, P2, and P3 were assigned through peak clustering. The performance of the proposed method was validated through a comparison of the algorithm-defined peaks and those manually identified by experienced observers. RESULTS: The proposed algorithm successfully identified the three distinguishing peaks of the ICP with satisfactory accuracy (95.3%, 87.8%, and 87.5% for P1, P2, and P3, respectively), even from minimally filtered raw signals. CONCLUSION: The algorithm extracted the morphological features from both ABP and ICP recordings with high accuracy. SIGNIFICANCE: The ABP and ICP pulse waveforms can be simultaneously analyzed in real time using the proposed algorithm. The morphological features from these signals may aid the continuous care of patients with TBI.

Spectral analysis of intracranial pressure: Is it helpful in the assessment of shunt functioning in-vivo?

OBJECTIVE: Shunt failure is common in hydrocephalic patients. The cerebrospinal fluid (CSF) infusion test enables the assessment of CSF absorption capacity, which is represented by the resistance to CSF outflow (ROUT) However, shunt failure may not only affect the CSF absorption capacity but also the intracranial compliance or compensatory properties. Spectral analysis of the ICP signal obtained during the infusion test may enable the comprehensive assessment of the overall deterioration caused by shunt failure. MATERIAL AND METHODS: A total of 121 hydrocephalic shunted patients underwent the infusion test with continuous intracranial pressure (ICP) and arterial blood pressure (ABP) recording. The maximum amplitudes of three major frequency bandwidths (0.2-2.6, 2.6-4.0 and 4.0-15 Hz, respectively) were calculated from the ICP. Statistical analyses were conducted to identify factors significantly associated with shunt failure, to construct an index (i.e., the shunt response parameter, SRP) for detecting shunt failure, and to define thresholds for ROUT and SRP. RESULTS: The ROUT threshold for detecting shunt failure was 7.59 mmHg/ml/min, and this threshold showed an accuracy of 82.64%. All spectral parameters were found to be significantly associated with shunt patency (p<0.05). The SRP exhibited significantly better accuracy than ROUT in detecting shunt failure (91.74%). CONCLUSION: The hydrodynamic assessment of shunted patients enhanced by spectral analysis during the infusion test detected shunt failure with high accuracy. Although further validation is needed, the SRP exhibited promising results.