Longitudinal Segmentation Research Articles

Development of Detector Control System (DCS) for forward hadron calorimeters in the BM@N and the MPD experiments

The BM@N and the MPD experiments at NICA facility (Dubna, Russia) will use the forward hadron calorimeters (FHCal) for centrality and reaction plane determination in the heavy ion collisions. The FHCal is a modular lead/scintillator sampling calorimeter with longitudinal segmentation. The light from each longitudinal section in module is detected with one silicon photomultiplier (MPPC) with sensitive area of 3x3mm2. In total, the FHCal at the BM@N experiment has 438 readout channels, while two MPD calorimeter arms have 616 readout channels. This article shows the DCS design with hardware and software development for these calorimeters. The DCS will be used to control and correct the bias voltages with temperatures of MPPCs. An integration into the main DCS systems for both experiments are discussed.

NIMG-55. A QUANTITATIVE ANALYSIS OF BRAIN VOLUME DYNAMICS IN PCNSL PATIENTS TREATED WITH HIGH-DOSE METHOTREXATE-BASED THERAPY

Abstract BACKGROUND Primary CNS lymphoma (PCNSL) is a rare, infiltrative disease. High-dose methotrexate (HD-MTX) is the backbone of induction regimens for PCNSL. While MTX-associated white matter changes are well-described, treatment-related brain volume loss is much less understood, especially in radiotherapy-naïve cohorts. Here, we aimed to longitudinally quantify the rates of brain volume loss in PCNSL patients treated with HD-MTX. SUBJECTS/METHODS We included 12 radiotherapy-naïve patients (age mean±SD 61±15y, range 37-84y, 9F) with histopathologically confirmed PCNSL who received HD-MTX induction (mean±SD 12±4 cycles, range 8-18) +/-rituximab. MRIs were collected from within 1 month of HD-MTX initiation until the end of follow-up (mean±SD: 3.7±2.9y). Longitudinal whole-brain segmentation was performed on FLAIR images using the Sequence Adaptive Multimodal Segmentation tool of FreeSurfer. Brain volumes were normalized to the initial scan, white matter lesion volumes were normalized to cerebral volume (nWML). RESULTS The average rate of cerebral volume change was -2.1±1.9%/year. Half of patients showed marked cerebral volume loss in the first year (-5.6±1.4% vs. -2.0±1.4%; n=10; p=0.003) with the most prominent change occurring within 6-months of treatment initiation (-4.2±1.7% vs. -0.5±1.6; n=12; p=0.004). Cerebral volume loss reached a plateau after the 1-year mark in both groups (0.3±0.8%/year vs. 1.4±3.3%/year; n=8; p=0.4). Patients younger than 61 years exhibited markedly higher rates of cerebral volume loss (-6.2±1.1%/year vs. 2.4±1.5%/year p=0.003), which was corroborated by strong inverse correlation between age and cerebral volume loss (Pearson’s r=-0.82, p=0.004). Neither the cerebellar volume, nor the nWML load correlated with age. CONCLUSION In the present cohort, brain volume loss was approximately four-fold higher than what is described in the healthy aging. Younger patients treated with HD-MTX exhibited more severe cerebral volume loss, which may be due to higher initial brain volume reserve. Detailed analyses of a larger sample are underway.

New perspectives on segmented crystal calorimeters for future colliders

Crystal calorimeters have a long history of pushing the frontier on high-resolution electromagnetic (EM) calorimetry for photons and electrons. We explore in this paper major innovations in collider detector performance that can be achieved with crystal calorimetry when longitudinal segmentation and dual-readout capabilities are combined with a new high EM resolution approach to Particle Flow in multi-jet events, such as e+e+→ HZ events in all-hadronic final-states at Higgs factories. We demonstrate a new technique for pre-processing π0 momenta through combinatoric di-photon pairing in advance of applying jet algorithms. This procedure significantly reduces π0 photon splitting across jets in multi-jet events. The correct photon-to-jet assignment efficiency improves by a factor of about 3 when the EM resolution is improved from 15 to 3%/√E. In addition, the technique of bremsstrahlung photon recovery significantly improves electron momentum measurements. A high EM resolution calorimeter increases the Z boson recoil mass resolution in Higgstrahlung events for decays into electron pairs to 80% of that for muon pairs. We present the design and optimization of a highly segmented crystal detector concept that achieves the required energy resolution of 3%/√E, and a time resolution better than 30 ps providing exceptional particle identification capabilities. We demonstrate that, contrary to previous detector designs that suffered from large neutral hadron resolution degradation from one interaction length of crystals in front of a sampling hadron calorimeter, the implementation of dual-readout on crystals permits to achieve a resolution better than 30%/√E⊕ 2% for neutral hadrons. Our studies find that the integration of crystal calorimetry into future Higgs factory collider detectors can open new perspectives by yielding the highest level of combined EM and neutral hadron resolution in the PFA paradigm.

Study of the hadron calorimeters response for CBM and BM@N experiments at hadron beams

The results of beam tests of the hadron calorimeter with transverse and longitudinal segmentation and with the micropixel photodetectors light readout performed at CERN T9 and T10 beamlines in proton momentum range 2-10 GeV/c are shown. The new signal processing technique based on the waveform fitting of calorimeter signals using the Prony least squares method is proposed. This technique allows to identify weak signals comparable to the level of electronic noise, which is important for performing a muon calibration of calorimeter sections. For the energy calibration of the hadron calorimeter sections with cosmic muons, a new approach that uses the reconstruction of the muon track in the calorimeter is proposed.

The readout system of the CBM Projectile Spectator Detector at FAIR

The Projectile Spectator Detector (PSD) is a sampling lead/scintillator forward hadron calorimeter with transverse and longitudinal segmentation and with MPPC photodetectors light readout. The PSD will be used at the Compressed Baryonic Matter (CBM) experiment at FAIR to measure the nucleus-nucleus collision centrality and orientation of the reaction plane. The CBM experiment will use a free-streaming data acquisition system (DAQ), which requires a coordinated time stamping of data in all sub-systems. To test prototypes of the CBM sub-detectors, front-end electronics (FEE) and readout electronics at high heavy ion beam rates the “mini CBM” (mCBM) installation has been assembled at SIS18 accelerator in GSI, Darmstadt, Germany in the framework of the FAIR Phase-0 program. A single “mini PSD” (mPSD) module has been integrated into the mCBM experiment. Details of the mPSD readout electronics and the first results of the data acquisition, processing and transmission within the common synchronized mCBM data transport system taken during the first beam tests are shown.

Methods for centrality determination in nucleus-nucleus collisions with forward hadron calorimeters at the BM@N experiment

The BM@N (Baryonic Matter at Nuclotron) is the fixed target experiment at the Nuclotron (JINR, Dubna, Russia) accelerator complex. The main goal of the BM@N experiment is studying properties of dense nuclear matter produced in the ion-ion collisions at beam energies up to 4.5 AGeV. First experiments were done for carbon, argon and krypton ions beams with C, Al, Cu and Pb targets. The methods of central events selection on event-by-event base with ZDC (Zero Degree Calorimeter) for these reactions are presented. The new Forward Hadron Calorimeter (FHCal) with transverse and longitudinal segmentation will be used to measure the collision centrality in heavy ion experiments after the BM@N upgrade. The FHCal has the beam hole in the center due to expected high beam intensities. New forward quartz hodoscope is been developed to be placed in the beam hole to measure the charge of fragments. New methods of centrality determination with the FHCal and forward quartz hodoscope are discussed.

Silicon Photomultipliers for the decay tunnel instrumentation of the ENUBET neutrino beam

The goal of the ENUBET project is to demonstrate that a precision of ∼1% on measurement of the absolute neutrino cross section at GeV scale can be achieved by monitoring the positron production in the decay tunnel coming from the three-body semileptonic decays of kaons. The baseline option for the tunnel instrumentation employs a fine-grained shashlik calorimeter with a 4.3 X0 longitudinal segmentation to separate positrons and pions coming from other decay modes of kaons. The system is complemented by rings of plastic scintillator doublets below the calorimeter acting as a photon veto to suppress the π0 background and to provide timing informations. SiPMs instrumenting the detector will be exposed to sizeble amounts of neutrons arising in hadronic showers. In order to reproduce such a working environment, SiPMs with different cell size (from 12 to 20 μm) were irradiated at the INFN-LNL CN Van Der Graaf with neutron fluences up to 2×1011 n/cm2 (1 MeV-eq.). The exposed light sensors were characterized in terms of I–V curves at different irradiation levels, and their response tested by exposing a prototype on beam at CERN. In this contribution we will report the results of the described tests on SiPMs, together with the advances in their integration with the ENUBET detectors.

Towards Optimising MRI Characterisation of Tissue (TOMCAT) Dataset including all Longitudinal Automatic Segmentation of Hippocampal Subfields (LASHiS) data.

Seven healthy participants were scanned using a Siemens Magnetom 7 Tesla (T) whole-body research MRI scanner (Siemens Healthcare, Erlangen, Germany). The first scan session was acquired in 2016 (time point one), the second and third session in 2019 (time point two and three, respectively) with the third session acquired 45 min following the second as a scan-rescan condition. The following scans were acquired for all time points: structural T1 weighted (T1w) MP2RAGE, high in-plane resolution Turbo-Spin Echo (TSE) dedicated for hippocampus subfield segmentation. The data were used in three projects to date, for more insight see: 1) Non-linear realignment for Turbo-Spin Echo retrospective motion correction and hippocampus segmentation improvement [1] 2) Longitudinal Automatic Segmentation of Hippocampal Subfields (LASHiS) using multi-contrast MRI [2]. 3) The challenge of bias-free coil combination for quantitative susceptibility mapping at ultra-high field [3]. Data were converted from DICOM to nifti format following the Brain Imaging Data Structure (BIDS) [4]. Data were analysed for the accompanying manuscript “Longitudinal Automatic Segmentation of Hippocampal Subfields (LASHiS) using multi-contrast MRI” including test-retest reliability and longitudinal Bayesian Linear Mixed Effects (LME) modelling.

Precision timing calorimetry with the CMS HGCAL

The existing CMS endcap calorimeters will be replaced with a High Granularity Calorimeter (HGCAL) for operation at the High Luminosity (HL) LHC . Radiation hardness and excellent physics performance will be achieved by utilising silicon pad sensors and SiPM-on-scintillator tiles with high longitudinal and transverse segmentation. One of the major challenges of the HL-LHC will be the high pileup environment, with interaction vertices spread not only in position, but also in time. In order to efficiently reject particles originating from pileup, precision timing information of the order of 30 ps will be of great benefit. In order to meet such performance goals, the HGCAL will provide timing measurements for individual hits with signals above 12 fC (equivalent to 3–10 MIPs), such that clusters resulting from particles with pT > 5 GeV should have a timing resolution better than 30 ps. Given the complexity and size of the system, this poses a particular challenge to the readout electronics as well as to the calibration and reconstruction procedures. We present the challenges for the front-end electronics design, results from prototype tests in laboratory and beam environments, as well as anticipated timing performance from simulation.

Calorimetry in Particle Physics, and the CMS High-Granularity Calorimeter

An overview of electromagnetic and hadronic shower development is presented, including how they lead to the design of modern calorimeters for energy measurements of high-energy particles. The differences between sampling and homogeneous calorimeters are explored, with the pros and cons of each being presented. Some real-world examples are given, focusing on state-of-the-art detectors designed and operating at the Large Hadron Collider. Methods to increase the information content from future calorimeters are given, and the challenges that these new calorimeters pose to detector physicists and engineers are discussed. Advances in large-area highly-segmented detectors are providing possibilities for high-granu-larity calorimetry. The CMS HGCAL, being designed to replace the existing CMS endcap calorimeters for the HL-LHC era, is one example. It is a sampling calorimeter, featuring unprecedented transverse and longitudinal readout segmentation for both electromagnetic (CE-E) and hadronic (CE-H) compartments. This will facilitate particle-flow calorimetry, where the fine structure of showers can be measured and used to enhance pileup rejection and particle identification, whilst still achieving good energy resolution. The CE-E and a large fraction of CE-H will use hexagonal silicon sensors as active detector material. The lower-radiation environment will be instrumented with scintillator tiles with on-tile SiPM readout. These concepts borrow heavily from designs produced by the CALICE collaboration but the design of such a detector at a hadron collider is considerably more challenging than at the linear colliders.

Performance of CMS High Granularity Calorimeter prototypes in beam test experiments

The present electromagnetic and hadronic calorimeters of the CMS experiment will be upgraded to cope with the harsh radiation environment and pileup conditions posed by the high luminosity operations of the LHC (HL-LHC) expected to start in 2027. The CMS collaboration has opted for a sampling calorimeter in the endcap sections of the detector, based on silicon and scintillator technologies, with unprecedented transverse and longitudinal segmentation to facilitate particle identification, particle-flow reconstruction and pileup rejection. As part of the ongoing development and testing phase of this High Granularity Calorimeter (HGCAL), prototypes of both the silicon and scintillator based calorimeter sections have been tested in 2018 in beams at CERN. We report on the performance of the prototype detectors in terms of stability of noise and pedestals, calibration with minimum-ionizing-particles (MIPs), and energy linearity and resolution for electrons and pions. We compare the measurements with a detailed GEANT4-based simulation.

Methods of signal processing and cosmic muon calibration for the BM@N sampling lead/scintillator hadron calorimeter

Recently developed new forward lead/scintillator sampling hadron calorimeters with transverse and longitudinal segmentation will be used in the upgraded BM@N (Baryonic Matter at Nuclotron) experiment at the Nuclotron-NICA acceleration complex in JINR (Dubna). These calorimeters are very important for measurements of centrality and the reaction plane orientation in heavy-ion collisions. Digitizing the analog signals in a wide dynamic range with sampling ADC leads to the fluctuations of the measured charge. The procedure of signal processing based on the Prony least squares fit method has been developed to improve the response requiring minimum machine time. In addition, fitting the signals with a known function allows one to select effectively weak signals at the level of electronics noise, that is important for performing a calibration of calorimeter sections with cosmic muons for the BM@N experiment. The same procedure will be used for hadron calorimeter in CBM experiment at FAIR. New methods of signal processing and cosmic muon tracking for forward hadron calorimeters will be presented.

New forward hadron calorimeter and hodoscope for the BM@N heavy ions experiment

New forward hadron calorimeter with transverse and longitudinal segmentation has been developed and constructed for the upgraded fixed target BM@N experiment at JINR, Dubna. The main purpose of this calorimeter is to provide event-by-event centrality and reaction plane orientation measurements in nucleus-nucleus collisions. The design of the hadron calorimeter composed of sampling lead/scintillator modules with a beam hole in the center is discussed. The light collection from longitudinal sections in modules is provided by Wave Length Shifting (WLS) fibers embedded in scintillator plates. Micro-pixel photodetectors (Hamamatsu MPPCs) are used for light detection from each longitudinal section of modules. The measured light yield is about 50 photoelectrons per section for MIPs. Sampling ADCs are used for signal readout in the calorimeter. The performance of the FHCAL is discussed. For centrality measurement additional forward quartz hodoscope is developed to measure charge of fragments in the FHCAL beam hole. The method of collision centrality determination is discussed.

The Phase II upgrade of the LHCb calorimeter system

The Phase 2 LHCb Upgrade, which is now being developed by the LHCb Collaboration, is supposed to make LHCb able to work at high luminosity, up to 2 ⋅ 1034/cm2/s. The plan is to collect ∼300 fb−1 of data in a few years. The Phase 2 Upgrade will require a major revision of the LHCb Calorimeter system. The increased instantaneous and integrated luminosity will result in very high particle density and radiation doses in the areas close to the beam pipe. In these conditions, ECAL has to provide high quality energy and position measurements for electromagnetic showers, as well as separation of two nearby developing showers. Another requirement for the LHCb Phase 2 Upgrade ECAL is the ability to measure the time of arrival of the photon or electron with an accuracy of few tens of picoseconds. At high luminosity, such time measurement is a powerful tool helping to correctly assign electromagnetic showers to primary vertices. The choice for the central part of ECAL can be a sampling calorimeter with dense tungsten-based converter and radiation hard crystal scintillator. Concerning the particular structure type, it was found that the Spaghetti Calorimeter (SPACAL) concept meets all the requirements, including limits on radiation degradation, if its active elements are made of GAGG:Ce or YAG:Ce scintillators. The peripheral areas with moderate radiation doses can be instrumented with calorimeter modules based on plastic scintillator. A R&D campaign was started to optimize the Upgrade 2 ECAL structure. It includes studies of scintillating materials, beam test studies of the performance of various ECAL module prototypes, and simulation studies to find the optimal detector layout, longitudinal segmentation and granularity. In this paper we present the R&D results and the current status of the LHCb Calorimeter upgrade.

Longitudinal Automatic Segmentation of Hippocampal Subfields (LASHiS) using multi-contrast MRI

The volumetric and morphometric examination of hippocampus formation subfields in a longitudinal manner using in vivo MRI could lead to more sensitive biomarkers for neuropsychiatric disorders and diseases including Alzheimer’s disease, as the anatomical subregions are functionally specialised. Longitudinal processing allows for increased sensitivity due to reduced confounds of inter-subject variability and higher effect-sensitivity than cross-sectional designs. We examined the performance of a new longitudinal pipeline (Longitudinal Automatic Segmentation of Hippocampus Subfields [LASHiS]) against three freely available, published approaches. LASHiS automatically segments hippocampus formation subfields by propagating labels from cross-sectionally labelled time point scans using joint-label fusion to a non-linearly realigned ‘single subject template’, where image segmentation occurs free of bias to any individual time point. Our pipeline measures tissue characteristics available in in vivo high-resolution MRI scans, at both clinical (3 ​T) and ultra-high field strength (7 ​T) and differs from previous longitudinal segmentation pipelines in that it leverages multi-contrast information in the segmentation process. LASHiS produces robust and reliable automatic multi-contrast segmentations of hippocampus formation subfields, as measured by higher volume similarity coefficients and Dice coefficients for test-retest reliability and robust longitudinal Bayesian Linear Mixed Effects results at 7 ​T, while showing sound results at 3 ​T. All code for this project including the automatic pipeline is available at https://github.com/CAIsr/LASHiS.

STRUCTURAL-KINEMATIC EVOLUTION OF THE EARTH'S CRUST OF THE DNIEPER-DONETS PALEORIFT<br/> Part 2. Systematic organization of post-rift structural transformations

At the stages of platform tectonic activation rock geomasses in riftogenic structures acquires the ability to horizontal displacement due to internal structural and material transformations under the influence of reid deformation in stress metamorphism zones. The article highlights the results of tectonophysical studies of post-rift complications of the architecture of the Dnieper-Donets paleorift. The structural manifestation of reid dynamic tectonics in the geological structure of the sedimentary complexes of the paleorift platform cover is strike-slip transformations of its primary linear elements — fault systems, tectonic segments, geoblocks, and anticlinal structure zones.Using the original method of reconstruction of tectonic stress and strain fields and tectonophysical analysis of geostructures, based on an analysis of anomalies of potential geophysical fields and vertical amplitudes of neotectonic movements, the territorial distribution, tectonic position and natural kinematic mechanisms of the formation of secondary deformation structures of tectonic flow, linear and areal concentration zones of reid deformations of subregional and zonal scales, size from hundreds of meters to tens of kilometers.The natural mechanisms of reid deformation are caused by horizontal-plane movements of structural elements, rocky ensembles of discontinuities and local structures of tectonic flow dynamically conjugated with them. The usual tectonic process of reid dynamic tectonics is the collision warping of the horizons of the sedimentary sequence, which causes deformation structure formation. Under its influence, structural waves of deformations of horizontal-plane tectonic flow were formed in the sedimentary cover of the paleorift, which determine the longitudinal tectonic segmentation of the geostructure and the main features of the system organization of the latest paleorift architecture.

Fully automated longitudinal segmentation of new or enlarged multiple sclerosis lesions using 3D convolutional neural networks

The quantification of new or enlarged lesions from follow-up MRI scans is an important surrogate of clinical disease activity in patients with multiple sclerosis (MS). Not only is manual segmentation time consuming, but inter-rater variability is high. Currently, only a few fully automated methods are available. We address this gap in the field by employing a 3D convolutional neural network (CNN) with encoder-decoder architecture for fully automatic longitudinal lesion segmentation.Input data consist of two fluid attenuated inversion recovery (FLAIR) images (baseline and follow-up) per patient. Each image is entered into the encoder and the feature maps are concatenated and then fed into the decoder. The output is a 3D mask indicating new or enlarged lesions (compared to the baseline scan). The proposed method was trained on 1809 single point and 1444 longitudinal patient data sets and then validated on 185 independent longitudinal data sets from two different scanners. From the two validation data sets, manual segmentations were available from three experienced raters, respectively. The performance of the proposed method was compared to the open source Lesion Segmentation Toolbox (LST), which is a current state-of-art longitudinal lesion segmentation method.The mean lesion-wise inter-rater sensitivity was 62%, while the mean inter-rater number of false positive (FP) findings was 0.41 lesions per case. The two validated algorithms showed a mean sensitivity of 60% (CNN), 46% (LST) and a mean FP of 0.48 (CNN), 1.86 (LST) per case. Sensitivity and number of FP were not significantly different (p < 0.05) between the CNN and manual raters.New or enlarged lesions counted by the CNN algorithm appeared to be comparable with manual expert ratings. The proposed algorithm seems to outperform currently available approaches, particularly LST. The high inter-rater variability in case of manual segmentation indicates the complexity of identifying new or enlarged lesions. An automated CNN-based approach can quickly provide an independent and deterministic assessment of new or enlarged lesions from baseline to follow-up scans with acceptable reliability.

Cross-organ, cross-modality transfer learning: feasibility study for segmentation and classification.

We conducted two analyses by comparing the transferability of a traditionally transfer-learned CNN (TL) to that of a CNN fine-tuned with an unrelated set of medical images (mammograms in this study) first and then fine-tuned a second time using TL, which we call the cross-organ, cross-modality transfer learned (XTL) network, on 1) multiple sclerosis (MS) segmentation of brain magnetic resonance (MR) images and 2) tumor malignancy classification of multi-parametric prostate MR images. We used 2133 screening mammograms and two public challenge datasets (longitudinal MS lesion segmentation and ProstateX) as intermediate and target datasets for XTL, respectively. We used two CNN architectures as basis networks for each analysis and fine-tuned it to match the target image types (volumetric) and tasks (segmentation and classification). We evaluated the XTL networks against the traditional TL networks using Dice coefficient and AUC as figure of merits for each analysis, respectively. For the segmentation test, XTL networks outperformed TL networks in terms of Dice coefficient (Dice coefficients of 0.72 vs [0.70 - 0.71] with p-value < 0.0001 in differences). For the classification test, XTL networks (AUCs = 0.77 - 0.80) outperformed TL networks (AUC = 0.73 - 0.75). The difference in the AUCs (AUCdiff = 0.045 - 0.047) was statistically significant (p-value < 0.03). We showed XTL using mammograms improves the network performance compared to traditional TL, despite the difference in image characteristics (x-ray vs. MRI and 2D vs. 3D) and imaging tasks (classification vs. segmentation for one of the tasks).

NIMG-63. LONGITUDINAL ANALYSIS OF OLIGODENDROGLIOMA GROWTH PATTERN REVEALED SPATIAL HETEROGENEITY AND DIVERSE TREATMENT RESPONSE

Abstract Oligodendroglioma are defined by a distinct molecular phenotype marked by 1p19q co-deletion and simultaneous presence of an IDH1/2 mutation. These tumors showed a favorable clinical course and long-term survival of around 15 years. Due to the long course of the disease, prospective studies to determine the effectiveness of different therapeutic strategies are difficult, since the percentage of patients with multiple therapies is high. Here we report a computational approach to map the longitudinal growth pattern, to quantify the effect of therapies on tumor growth and to identify similarities and spatial heterogeneity of oligodendroglioma growth. In our study, we included a cohort of 44 histopathologically and molecularly stratified oligodendrogliomas WHO°II (n=23) and WHO°III (n=21). We started our investigation with the longitudinal tumor segmentation. All volumetric data were pinpointed to the times of tumor therapy within all patients. Next, we extracted first-order features of tumor growth and response to chemo- or radiotherapy as well as resection, resulting in a total number of 98 features. An unsupervised cluster was used to identify similarities between patients, which revealed 3 subgroups. The first subgroup contained patients with predominantly frontal oligodendrogliomas marked by increased response to radiotherapy. The second subgroup included temporal oligodendrogliomas with high response rate to PC/PCV chemotherapy and flagged by epilepsy. The third group was heterogeneous with varying growth behaviors. A survival analysis showed a better separation between low- and high-risk patients based on the growth pattern model, in contrast to the WHO grading system. Taken together, our analysis revealed a novel classification of oligodendroglioma based on the longitudinal growth pattern and therapeutical response. We also detected a spatial difference between frontally or temporally localized oligodendrogliomas. We plan to further investigate molecular data that explain these spatial differences, which also may uncover novel therapeutic strategies.

A novel machine learning approach for evaluation of public policies: An application in relation to the performance of university researchers

Research has become the main reference point for academic life in modern universities. Research incentives have been a controversial issue, because of the difficulty of identifying who are the main beneficiaries and what are the long-term effects. Still, new policies including financial incentives have been adopted to increase the research output at all possible levels. Little literature has been devoted to the response to those incentives. To bridge this gap, we carry out our analysis with data of a six years program developed in Madrid (Spain). Instead of using a traditional econometric approach, we design a machine learning multilevel model to discover on whom, when, and for how long those policies have an effect. The empirical model consists of an automated nested longitudinal clustering (ANLC) performed in two stages. Firstly, it performs a stratification of academics, and secondly, it performs a longitudinal segmentation for each group. The second part considers the researchers’ sociodemographic, academic information and the evolution of their performance over time in the form of the annual percentage variation of their marks over the period. The new methodology, whose robustness is tested with a multilayer perceptron artificial neural network with a back-propagation learning algorithm, shows that tenure track researchers present a better response to incentives than tenured researches, and also that gender plays an important role in academia.These discoveries are relevant to administrations and universities for understanding the productivity of academics working under long-term incentive-based programs, the drawbacks and the inequalities for maximizing the generation of knowledge.