Cross-platform hyperparameter optimization for machine learning interatomic potentials.

Machine-learning (ML)-based interatomic potentials are increasingly popular in material modeling, enabling highly accurate simulations with thousands and millions of atoms. However, the performance of machine-learned potentials depends strongly on the choice of hyperparameters-that is, of those parameters that are set before the model encounters data. This problem is particularly acute where hyperparameters have no intuitive physical interpretation and where the corresponding optimization space is large. Here, we describe an openly available Python package that facilitates hyperparameter optimization across different ML potential fitting frameworks. We discuss methodological aspects relating to the optimization itself and to the selection of validation data, and we show example applications. We expect this package to become part of a wider computational framework to speed up the mainstream adaptation of ML potentials in the physical sciences.

Long-term results of stent graft treatment of subclavian artery injuries: management of choice for stable patients?

BACKGROUND: The management of penetrating subclavian artery injuries poses a formidable surgical challenge. The feasibility of stent graft repair is already established. General use of this modality is not widely accepted due to concerns regarding the long-term outcome in a generally young patient population. We review our stent graft experience to examine long-term outcomes. METHODS: All patients with penetrating subclavian artery injuries were evaluated for stent graft repair. Patients were excluded when hemodynamically unstable or unsuitable on other clinical and angiographic grounds. Patients were followed prospectively for early (<30 days) and late (>30 days) complications. Clinical and telephone evaluation, Doppler pressures, duplex Doppler, and angiography (when indicated), were used to asses patients at follow-up. Outcomes were recorded as technical success of procedure, graft patency, arm claudication, limb loss, the need for open surgical repair, the presence or absence of other complications, and death. RESULTS: Fifty-seven patients underwent stent graft treatment during the 10-year period. Mean age was 34, and 91% were men. There were 53 stab wounds and four gunshot injuries. Pathology included false aneurysms (n = 42), arteriovenous fistula (n = 12), and three arterial occlusions. Early complications: One patient (2%) had a femoral puncture site injury which was managed with open surgical repair. One patient died early due to multiple organ failure related to concomitant injuries. Three patients (5%) presented with graft occlusion and nonlimb threatening ischemia in the first week after treatment. All three patients were managed successfully with a second endovascular intervention. Late complications: Twenty-five (44%) of the 57 patients with subclavian artery injuries were followed-up with a mean duration of 48 months. Two patients died as a result of fatal stab wounds months after their first injuries. Five patients (20%) and three patients (12%) presented with angiographically significant stenosis and occlusions, respectively. The stenotic lesions were successfully managed with endovascular intervention, and the occluded lesions were managed conservatively. No patient experienced life or limb loss or any incapacitating symptoms at the end of the study period. There was no need for conversion to open surgery. CONCLUSIONS: This study has reaffirmed the feasibility and safety of stent graft repair in treating stable patients with selected penetrating subclavian artery injuries. The results of this study also confirmed acceptable long-term follow-up without any limb or life threatening complications. We conclude that endovascular repair should be considered the first choice of treatment in stable patients with subclavian artery injuries.

Neurologic outcome after penetrating extracranial arterial trauma.

PURPOSE: We undertook this study to determine factors that adversely affect outcome in patients with penetrating injury to the extracranial cerebral vasculature. Patients and methods Medical records were reviewed for all patients who had undergone surgical intervention to treat penetrating injury to the extracranial cerebral arteries between January 1989 and December 1999. Forensic autopsy findings were also reviewed for all patients who died as a result of their injury. RESULTS: One hundred fifty-one patients with injury to the brachiocephalic artery (n = 21), common carotid artery (n = 98), or internal carotid artery (n = 32) were identified. Overall mortality was 21.2%, and stroke rate in surviving patients was 15.1%. Twenty-five of 32 deaths (78.1%) were stroke-related. Brachiocephalic artery injury was associated with the highest mortality (38.1%), and survivor stroke rate was highest in patients with internal carotid injuries (22.7%). Hemodynamic instability at presentation led to both higher mortality (30.7%) and stroke rate (19.2%). Preoperative angiography did not influence mortality or stroke rate in hemodynamically stable patients. Procedural mortality associated with arterial ligation was 45% (9 of 20 patients), and no surviving patient experienced a change in pre-ligation neurologic state. Nine patients remained neurologically intact after ligation, and 2 patients with preoperative localized neurologic deficit were unchanged postoperatively. In 131 patients, mortality after arterial repair was 17.6%, and in 5 surviving patients (5.4%) an ischemic neurologic deficit developed. Twelve of 15 surviving patients (80%) with preoperative neurologic deficit who underwent arterial repair had improved neurologic status. Cerebral infarcts were confirmed at autopsy in 23 patients; 18 infarcts were ischemic (10, repair; 8, ligation), and 5 infarcts were hemorrhagic (all, repair). No factor was identified that was predictive of ischemic versus hemorrhagic infarction in patients undergoing repair. CONCLUSIONS: The presence of hypovolemic shock, internal carotid artery injury, complete vessel transection, and arterial ligation are associated with unfavorable outcome. Penetrating injury to the brachiocephalic, common carotid, or internal carotid artery should be repaired rather than ligated when technically possible. Subsequent ischemic or hemorrhagic cerebral infarction is unpredictable, but overall outcome is superior to that with ligation of the injured artery.

Duodenal injuries: surgical management adapted to circumstances.

OBJECTIVES: To evaluate the results of an evolving policy of primary repair of duodenal injuries, when considered feasible and safe, complemented by pyloric exclusion if any doubt as to the integrity of the repair existed. PATIENTS AND METHODS: A prospective analysis of all patients with surgically identified duodenal injuries treated at a single institution over a 3-year period. RESULTS: In 30 patients studied, there were four deaths (13%) and an overall complication rate of 47%. There were 68 associated intra-abdominal injuries in 29 patients. Primary duodenorraphy was employed in 18 patients (60%), pyloric exclusion in 11 (37%) and pancreatoduodenectomy in one patient (3%). No failures of duodenal repair were recorded. CONCLUSION: Adverse results in patients with duodenal trauma are largely a reflection of frequent associated injuries and their consequences. Selective, liberal employment of pyloric exclusion, based on individualized intra-operative assessment, can minimize duodenum-related morbidity.