Resilient multi-agent RL: introducing DQ-RTS for distributed environments with data loss.

This paper proposes DQ-RTS, a novel decentralized Multi-Agent Reinforcement Learning algorithm designed to address challenges posed by non-ideal communication and a varying number of agents in distributed environments. DQ-RTS incorporates an optimized communication protocol to mitigate data loss between agents. A comparative analysis between DQ-RTS and its decentralized counterpart Q-RTS, or Q-learning for Real-Time Swarms, demonstrates the superior convergence speed of DQ-RTS, achieving a remarkable speed-up factor ranging from 1.6 to 2.7 in scenarios with non-ideal communication. Moreover, DQ-RTS exhibits robustness by maintaining performance even when the agent population fluctuates, making it well-suited for applications requiring adaptable agent numbers over time. Additionally, extensive experiments conducted on various benchmark tasks validate the scalability and effectiveness of DQ-RTS, further establishing its potential as a practical solution for resilient Multi-Agent Reinforcement Learning in dynamic distributed environments.

Sensing and Detection of Traffic Signs Using CNNs: An Assessment on Their Performance.

Traffic sign detection systems constitute a key component in trending real-world applications such as autonomous driving and driver safety and assistance. In recent years, many learning systems have been used to help detect traffic signs more accurately, such as ResNet, Vgg, Squeeznet, and DenseNet, but which of these systems can perform better than the others is debatable. They must be examined carefully and under the same conditions. To check the system under the same conditions, you must first have the same database structure. Moreover, the practice of training under the same number of epochs should be the same. Other points to consider are the language in which the coding operation was performed as well as the method of calling the training system, which should be the same. As a result, under these conditions, it can be said that the comparison between different education systems has been done under equal conditions, and the result of this analogy will be valid. In this article, traffic sign detection was done using AlexNet and XresNet 50 training methods, which had not been used until now. Then, with the implementation of ResNet 18, 34, and 50, DenseNet 121, 169, and 201, Vgg 16_bn and Vgg19_bn, AlexNet, SqueezeNet1_0, and SqueezeNet1_1 training methods under completely the same conditions. The results are compared with each other, and finally, the best ones for use in detecting traffic signs are introduced. The experimental results showed that, considering parameters train loss, valid loss, accuracy, error rate and Time, three types of CNN learning models Vgg 16_bn, Vgg19_bn and, AlexNet performed better for the intended purpose. As a result, these three types of learning models can be considered for further studies.

A pseudo-softmax function for hardware-based high speed image classification.

In this work a novel architecture, named pseudo-softmax, to compute an approximated form of the softmax function is presented. This architecture can be fruitfully used in the last layer of Neural Networks and Convolutional Neural Networks for classification tasks, and in Reinforcement Learning hardware accelerators to compute the Boltzmann action-selection policy. The proposed pseudo-softmax design, intended for efficient hardware implementation, exploits the typical integer quantization of hardware-based Neural Networks obtaining an accurate approximation of the result. In the paper, a detailed description of the architecture is given and an extensive analysis of the approximation error is performed by using both custom stimuli and real-world Convolutional Neural Networks inputs. The implementation results, based on CMOS standard-cell technology, compared to state-of-the-art architectures show reduced approximation errors.

Digital measurements of air leak flow and intrapleural pressures in the immediate postoperative period predict risk of prolonged air leak after pulmonary lobectomy.

BACKGROUND: The objective of this prospective observational study was to evaluate the association between the airflow and intrapleural pressures digitally recorded during the immediate postoperative period after lobectomy and their ability to predict the risk of subsequent prolonged air leak (PAL). METHODS: A total of 145 consecutive patients underwent pulmonary lobectomy in two centers. All patients were managed with the chest tube placed on suction (-20 cm H(2)O) until the morning of the first postoperative day. Measurement of airflow and maximum and minimum intrapleural pressures were recorded during the 6th postoperative hour using a digital chest drainage device. Logistic regression analysis validated by bootstrap was used to test independent association of variables with PAL (air leak>72 h). RESULTS: The mean air leak flow at the 6th postoperative hour was 86 ml min(-1) (0-1100 ml min(-1)). The mean maximum and minimum pleural pressures at the 6th postoperative hour were -11.4 cm H(2)O and -21.9 cm H(2)O, respectively. Logistic regression and bootstrap showed that the mean air leak flow (p=0.007) and the mean differential pleural pressure (ΔP: maximum-minimum intrapleural pressure) (p=0.02) at the 6th postoperative hour were reliably associated with PAL, independent of the effect of age, forced expiratory volume 1 (FEV1), chronic obstructive pulmonary disease (COPD) status, diffusing capacity of the lung for carbon monoxide (DLCO), side, and site of lobectomy. According to best cutoffs derived by receiver operating characteristic (ROC) analysis the following combinations showed incremental risk of PAL: ΔP<10+Flow<50: 4% (3/73); ΔP>10+Flow<50: 15% (5/33); ΔP<10+Flow>50: 36% (5/14); ΔP>10+Flow>50: 52% (13/25). CONCLUSIONS: The levels of both air leak flow and pleural pressure measured at the 6th postoperative hour are associated to a different extent with the duration of air leak. Interpretation of the data measured at an early time point by digital chest drainage systems allows estimation of the risk of subsequent PAL. In this way, digital devices may help to plan postoperative management to allow both safe and more accurate implementation of fast-tracking strategies.

Prospective external convergence evaluation of two different quality-of-life instruments in lung resection patients.

OBJECTIVES: The interpretation of studies on quality of life (QoL) after lung surgery is often difficult owing to the use of multiple instruments with inconsistent scales and metrics. Although a more standardized approach would be desirable, the most appropriate instrument to be used in this setting is still largely undefined. The aim of the study was to assess the respective ability of two validated QoL instruments (European Organization for Research and Treatment of Cancer (EORTC) QLQ-C30/L13 and Short Form (36) Health Survey (SF-36)) to detect perioperative changes in QoL of patients submitted to pulmonary resection for non-small-cell lung cancer (NSCLC). METHODS: A prospective study on 33 consecutive patients (May 2009-December 2009) was submitted to pulmonary resection. All patients completed both EORTC QLQ-C30 with lung module 13 and SF-36 pre- and postoperatively (3 months). Preoperative changes of all SF-36 and EORTC scales were assessed by using the Cohen's effect-size method. External convergence between different instruments (SF-36 vs EORTC) was assessed by measuring the correlation of scales evaluating the same concepts (physical, psychosocial, and emotional). The correlation coefficients between standardized perioperative changes (effect sizes) of objective functional parameters (forced expiratory volume in 1s (FEV1) and diffusion lung capacity for carbon monoxide (DLCO)) and SF-36 or EORTC scales were also investigated. RESULTS: A poor correlation (r < 0.5) was detected between most of the scales of the two instruments measuring the same QoL concepts, indicating that they may be complementary in investigating different aspects of QoL. Only the SF-36 and EORTC social functioning scales and the SF-36 mental health and EORTC emotional functioning scales had a correlation coefficient >0.5. In general, EORTC was more sensitive in detecting physical or emotional declines but was more conservative in detecting improvements. Both SF-36 and EORTC showed poor correlations (r < 0.5) between perioperative changes in QoL and FEV1 or DLCO, confirming that objective parameters cannot be surrogates to the subjective perception of QoL. In particular, there was a poor correlation between perceived changes in dyspnea and objective changes in FEV1 or DLCO. CONCLUSIONS: EORTC behaved similarly to SF-36 in assessing perioperative changes in generic QoL scales, but, with the use of its lung module, provided a more detailed evaluation of specific symptoms. For this reason, EORTC should be regarded as the instrument of choice for measuring QoL in the thoracic surgery setting.

Peak oxygen consumption measured during the stair-climbing test in lung resection candidates.

BACKGROUND: The stair-climbing test is commonly used in the preoperative evaluation of lung resection candidates, but it is difficult to standardize and provides little physiologic information on the performance. OBJECTIVE: To verify the association between the altitude and the V(O2peak) measured during the stair-climbing test. METHODS: 109 consecutive candidates for lung resection performed a symptom-limited stair-climbing test with direct breath-by-breath measurement of V(O2peak) by a portable gas analyzer. Stepwise logistic regression and bootstrap analyses were used to verify the association of several perioperative variables with a V(O2peak) <15 ml/kg/min. Subsequently, multiple regression analysis was also performed to develop an equation to estimate V(O2peak) from stair-climbing parameters and other patient-related variables. RESULTS: 56% of patients climbing <14 m had a V(O2peak) <15 ml/kg/min, whereas 98% of those climbing >22 m had a V(O2peak) >15 ml/kg/min. The altitude reached at stair-climbing test resulted in the only significant predictor of a V(O2peak) <15 ml/kg/min after logistic regression analysis. Multiple regression analysis yielded an equation to estimate V(O2peak) factoring altitude (p < 0.0001), speed of ascent (p = 0.005) and body mass index (p = 0.0008). CONCLUSIONS: There was an association between altitude and V(O2peak) measured during the stair-climbing test. Most of the patients climbing more than 22 m are able to generate high values of V(O2peak) and can proceed to surgery without any additional tests. All others need to be referred for a formal cardiopulmonary exercise test. In addition, we were able to generate an equation to estimate V(O2peak), which could assist in streamlining the preoperative workup and could be used across different settings to standardize this test.

Chest drainage suction decreases differential pleural pressure after upper lobectomy and has no effect after lower lobectomy.

BACKGROUND AND OBJECTIVE: To our knowledge, no reports have been published describing the effect of suction on pleural pressures after different types of lobectomy. Improving knowledge of pleural physiology in the postoperative period could lead to better postoperative care. The aim of this investigation is to evaluate the effect of postoperative suction on inspiratory, expiratory and differential pleural pressures after upper or lower lobectomy. METHODS: Records of intrapleural pressures from 24 lobectomy patients (operated on in two different institutions) were selected for study. All patients had normal preoperative pulmonary function tests (forced expiratory volume in 1s (FEV1) >80% and forced vital capacity (FVC)/FEV1 >70%), and neither postoperative air leak nor any other postoperative complication. We selected six cases of each type of lobectomy (right upper lobectomy (RUL), right lower lobectomy (RLL), left upper lobectomy (LUL) and left lower lobectomy (LLL)). In three cases of each group, no suction was indicated, while in the other three cases, chest tubes were placed under 15 cm H(2)O suction, according to the standard local perioperative care protocol in each participating centre. Inspiratory and expiratory pleural pressures were measured at 2-min intervals by an electronic device using a DigiVent (Millicore A.B., Sweden) suction chamber. Recording started 5-10h after closing the chest, and included 5 consecutive hours during the first postoperative night, with the patients at rest in 30-45 degrees sitting position. There was no evidence of pneumothorax during the recording time. The influence of lobectomy site (upper or lower) and suction on inspiratory, expiratory and differential pressures were evaluated by Student's t-tests. RESULTS: In the group of cases under no suction, upper lobectomy patients had larger differential pressures (22.6 in upper vs 11.5 cm H(2)O in lower lobectomy cases, p<0.001), differential pressure decreased in patients under suction (9.1 in upper vs 11.1 cm H(2)O in lower lobectomy cases, p<0.001) and the effect was mainly due to a less negative inspiratory pressure. CONCLUSION: Pleural suction leads to a large decrease of differential pleural pressure after upper pulmonary lobectomy. The influence of this finding on postoperative work of breathing in the early postoperative period remains to be investigated.

Evaluation of a new chest tube removal protocol using digital air leak monitoring after lobectomy: a prospective randomised trial.

BACKGROUND: The objective of this randomised trial was to assess the effectiveness of a new fast-track chest tube removal protocol taking advantage of digital monitoring of air leak compared to a traditional protocol using visual and subjective assessment of air leak (bubbles). METHODS: One hundred and sixty-six patients submitted to pulmonary lobectomy for lung cancer were randomised in two groups with different chest tube removal protocols: (1) in the new protocol, chest tube was removed based on digitally recorded measurements of air leak flow; (2) in the traditional protocol, the chest tube removal was based on an instantaneous assessment of air leak during daily rounds. The two groups were compared in terms of chest tube duration, hospital stay and costs. RESULTS: The two groups were well matched for several preoperative and operative variables. Compared to the traditional protocol, the new digital recording protocol showed mean reductions in chest tube duration (p=0.0007), hospital stay (p=0.007) of 0.9 day, and a mean cost saving of euro 476 per patient (p=0.008). In the new chest tube removal protocol, 51% of patients had their chest tube removed by the second postoperative day versus only 12% of those in the traditional protocol. CONCLUSIONS: The application of a chest tube removal protocol using a digital drainage unit featuring a continuous recording of air leak was safe and cost effective. Although future studies are warranted to confirm these results in other settings, the use of this new protocol is now routinely applied in our practice.

Short-term perioperative treatment with ambroxol reduces pulmonary complications and hospital costs after pulmonary lobectomy: a randomized trial.

OBJECTIVE: To assess in a randomized clinical trial the influence of perioperative short-term ambroxol administration on postoperative complications, hospital stay and costs after pulmonary lobectomy for lung cancer. METHODS: One hundred and forty consecutive patients undergoing lobectomy for lung cancer (April 2006-November 2007) were randomized in two groups. Group A (70 patients): ambroxol was administered by intravenous infusion in the context of the usual therapy on the day of operation and on the first 3 postoperative days (1000 mg/day). Group B (70 patients): fluid therapy only without ambroxol. Groups were compared in terms of occurrence of postoperative complications, length of stay and costs. RESULTS: There were no dropouts from either group and no complications related to treatment. The two groups were well matched for perioperative and operative variables. Compared to group B, group A (ambroxol) had a reduction of postoperative pulmonary complications (4 vs 13, 6% vs 19%, p=0.02), and unplanned ICU admission/readmission (1 vs 6, 1.4% vs 8.6%, p=0.1) rates. Moreover, the postoperative stay and costs were reduced by 2.5 days (5.6 vs 8.1, p=0.02) and 2765 Euro (2499 Euro vs 5264 Euro, p=0.04), respectively. CONCLUSIONS: Short-term perioperative treatment with ambroxol improved early outcome after lobectomy and may be used to implement fast-tracking policies and cut postoperative costs. Nevertheless, other independent trials are needed to verify the effect of this treatment in different settings.