Deep Regression Model Research Articles

Deep Learning-Assisted Compound Bioactivity Estimation Framework

Drug Discovery is a highly complicated process. On average, it takes six to twelve years to manufacture a new drug and have the product released in the market. It is of utmost importance to find methods that would accelerate the manufacturing process. This significant challenge in drug development can be addressed using deep learning techniques. The aim of this paper is to propose a deep learning-based framework that can help chemists examine compound biological activity in a more accurate manner. The proposed framework employs autoencoder for data representation of the compounds data, which is then classified using deep neural network followed by building a customized deep regression model to estimate an accurate value of the compound bioactivity. The proposed framework achieved an accuracy of 89% in autoencoder reconstruction error, 79.01% in classification, and MAE of 2.4 while predicting compound bioactivity using deep regression model.

An Adaptive Missing Data Restoration Method for UAV Confrontation Based on Deep Regression Model

Completing missions with autonomous decision-making unmanned aerial vehicles (UAV) is a development direction for future battlefields. UAV make decisions based on battlefield situation information collected by sensors and can quickly and accurately perform complex tasks such as path planning, cooperative reconnaissance, cooperative pursuit and attacks. Obtaining real-time situation information of enemy is the basis for realizing autonomous decision-making of the UAV. However, in practice, due to internal sensor failure or interference of enemy, the acquired situation information is prone to be missing, which affects the training and decision-making of autonomous UAV. In this paper, an adaptive missing situation data restoration method for UAV confrontation is proposed. The UAV confrontation situation data are acquired through JSBSim, an open-source UAV simulation platform. By fusing temporal convolutional network and long short-term memory sequences, we establish a deep regression method for missing data restoration and introduce an adaptive mechanism to reduce the training time of the restoration model in response to dynamic changes in the enemy’s strategy during UAV confrontation. In addition, we evaluate the reliability of the proposed method by comparing with different baseline models under different degrees of data missing conditions. The performance of our method is quantified by five metrics. The performance of our proposed method is better than the other benchmark algorithms. The experimental results show that the proposed method can solve the missing data restoration problem and provide reliable situation data while effectively reducing the training time of the restoration model.

Integrating the Capsule-like Smart Aggregate-Based EMI Technique with Deep Learning for Stress Assessment in Concrete.

This study presents a concrete stress monitoring method utilizing 1D CNN deep learning of raw electromechanical impedance (EMI) signals measured with a capsule-like smart aggregate (CSA) sensor. Firstly, the CSA-based EMI measurement technique is presented by depicting a prototype of the CSA sensor and a 2 degrees of freedom (2 DOFs) EMI model for the CSA sensor embedded in a concrete cylinder. Secondly, the 1D CNN deep regression model is designed to adapt raw EMI responses from the CSA sensor for estimating concrete stresses. Thirdly, a CSA-embedded cylindrical concrete structure is experimented with to acquire EMI responses under various compressive loading levels. Finally, the feasibility and robustness of the 1D CNN model are evaluated for noise-contaminated EMI data and untrained stress EMI cases.

Measurement method for live chicken shank length based on improved ResNet and fused multi-source information

Phenotypic parameters are crucial reference indicators in poultry breeding. However, the chickens shank length is still manually measured, which is time-consuming and labor-intensive. Additionally, the measurement results are difficult to unify due to the subjective factors of different individuals. To address this issue, this paper proposed a method for live chicken shank length measurement (SLM). It enriches chicken shank feature by fusing visible images and infrared images. The fusion images are then input into a deep regression model based on the improved ResNet. The measurement model used ResNet as its backbone and introduces Squeeze-and-Excitation (SE) blocks and a Spatial Pyramid Pooling (SPP) block, resulting in more precise and stable shank length measurements. The average coefficient of variation, average floating error, average standard deviation and Pearson correlation coefficient for shank length measurements using the fusion images are 0.21 %, 0.49 %, 0.181 mm and 0.996, respectively, compared with using single visible or infrared image, the accuracy and stability are obviously improved. That indicated combining deep learning model and fusion information, the SLM proposed in this paper can achieve a more precise, reliable and standardized measurement of live chicken shank length.

Abstract LB384: Artificial intelligence measures of tumor infiltrating lymphocytes predict colorectal cancer-specific and overall survival

Abstract Tumor infiltrating lymphocytes (TILs) per high-power field (HPF) as measured by expert pathologists is a useful, independent, validated prognostic marker in colorectal cancer (CRC). Artificial intelligence (AI) techniques of deep learning (DL) can predict TILs directly from routinely collected, digitized hematoxylin and eosin (H&E) pathology slides, but AI-predicted TILs have not yet been thoroughly validated as clinically relevant biomarker. Here we test whether AI-predicted TILs per HPF can be validated as an independent prognostic biomarker of CRC-specific survival in patients. We constructed an AI model to predict TILS per HPF using a weakly-supervised, transformer-based deep regression model, trained using 5-fold cross validation in a large dataset (n=1,738) and validated in a second independent dataset (n=223) of CRC cases. The model predicted TILs per HPF as a continuous variable, which was carried forward for survival analyses both as a continuous variable and dichotomized at a previously established a priori threshold of <2 vs. ≥2 TILs per HPF. Using longitudinal data from the population-based Molecular Epidemiology of Colorectal Cancer Study (median follow-up = 95 months), survival analyses were performed using non-parametric and Cox-proportional hazards models, with and without adjustment for age, sex, ethnicity, stage, and molecularly measured microsatellite instability (MSI). Two or more AI-predicted TILs per HPF were significantly associated with 5-year CRC-specific survival (p=0.0000037), 5-year overall survival (p=0.00021), and overall survival (p=0.000067). In a Cox proportional hazards model adjusting for age, sex, ethnicity, stage, and MSI, ≥2 AI-predicted TILs per HPF was significantly associated with improved 5-year CRC-specific survival, with a hazard ratio (HR) = 0.62, (95% confidence interval; 0.43, 0.91), p=0.01. Our new AI-driven deep learning model, which we call HopeSTIL, provides a highly efficient algorithm for analyzing digital pathology images of H&E sections of CRC to quantify TILs per HPF. HopeSTIL is a validated prognostic marker for 5-year CRC-specific survival that does not require a pathologist to manually count and score TILs per HPF. Citation Format: Stephen B. Gruber, Omar S. El Nahhas, Joseph D. Bonner, Joel K. Greenson, Daniel Schmolze, Lawrence Shaktah, Jonathan Salazar, Lorena Reynaga, Sidney Lindsey, Jenny Lu, Allen Mao, Victor Moreno, Stephani L. Schmit, Ya-Yu Tsai, Stanley R. Hamilton, Gad Rennert, Jakob N. Kather. Artificial intelligence measures of tumor infiltrating lymphocytes predict colorectal cancer-specific and overall survival [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(7_Suppl):Abstract nr LB384.

Abstract LB386: Weakly-supervised prediction of tumor infiltrating lymphocytes per high power field from colorectal cancer histopathology slides using regression transformers

Abstract Tumor infiltrating lymphocytes (TILs)-based biomarkers have emerged as a robust method for predicting treatment efficacy and outcome in colorectal cancer (CRC). However, the quantification of TILs remains a strenuous task for pathologists and is susceptible to inter-pathologist variations. Deep learning (DL) can predict cancer biomarkers directly from routine hematoxylin and eosin (H&E) pathology slides, enabling the automated and consistent quantification of TILs for clinical decision-making. Our study focuses on the prediction of TILs per high-power field (HPF) on two large cohorts of CRC patients. We developed a weakly-supervised, transformer-based deep regression model to predict TILs per HPF from routine H&E-stained histology slides. We trained the model using 5-fold cross-validation on a large cohort of CRC patients (n=1,738), and validated its performance on an external cohort of CRC patients (n=223). The model was evaluated using the Pearson’s correlation coefficient r. Moreover, pathologist-determined TILs per HPF was categorized into low (<2) and high (≥2), enabling evaluation via the area under the receiver operating characteristic curve (AUROC). Our model achieved a significant correlation coefficient of 0.60 (p<0.0001) for the predicted TILs per HPF in the holdout test sets of the internal cohort, and a significant correlation coefficient of 0.57 (p<0.0001) on the external validation cohort. Using the thresholds for low and high TILs per HPF, we found that the model accurately predicted the presence of high TILs per HPF in both the internal and external test cohorts, yielding AUROCs of 0.76 and 0.81, respectively. These findings underscore the efficacy of our deep learning model in predicting TILs per HPF from routine pathology slides. This approach holds promise for cost-effective, efficient and uniform quantification of TILs per HPF in CRC, potentially across various cancer types. Citation Format: Omar S. El Nahhas, Joseph D. Bonner, Joel K. Greenson, Daniel B. Schmolze, Lawrence Shaktah, Jonathan Salazar, Lorena Reynaga, Sidney Lindsey, Jenny Lu, Victor Moreno, Stephanie L. Schmit, Ya-Yu Tsai, Stanley R. Hamilton, Gad Rennert, Jakob N. Kather, Stephen B. Gruber. Weakly-supervised prediction of tumor infiltrating lymphocytes per high power field from colorectal cancer histopathology slides using regression transformers [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(7_Suppl):Abstract nr LB386.

Abstract 6191: Simultaneous prediction of tumor microenvironment biomarkers from pathology slides using multi-task deep regression

Abstract Understanding the Tumor Microenvironment (TME) is crucial for tailoring personalized cancer treatments and improving patient outcomes. Using deep learning, we can predict properties of the TME from standard Hematoxylin and Eosin (H&E) stained slides, including specific gene expression signatures. While past models were limited to single signature predictions, we explore the potential of multi-task deep learning to identify multiple biomarkers at once, allowing more complex relationships and interactions to be modeled simultaneously. Our study focuses on the prediction of expression patterns of tumor, stromal, and immune cells in breast cancer (BC). We developed a multi-task transformer-based deep regression model, trained in a weakly-supervised setting, to predict biomarkers in the TME from routine H&E-stained pathology slides of a BC cohort (n=876). The performance was evaluated using Pearson’s correlation coefficient (r) and spatial heatmaps generated via gradient-weighted class activation mapping. Our model achieved a significant correlation coefficient above 0.40 (p<0.0001) for all predicted TME biomarkers in the holdout test set (n=176). This includes the biomarkers lymphocyte infiltrating signature score (r=0.46, p<0.0001), tumor infiltrating lymphocyte regional fraction (r=0.48, p<0.0001), leukocyte fraction (r=0.43, p<0.0001), stromal fraction (r=0.42, p<0.0001), and tumor cell proliferation (r=0.47, p<0.0001). Notably, the model's heatmaps highlighted the pertinent cells and histological characteristics for its predictions. These data show that our deep learning model accurately predicts multiple TME biomarkers concurrently from routine pathology slides. This approach offers a promising avenue for cost-effective and efficient biomarker quantification in the TME, with potential applicability across various cancer types. Citation Format: Omar S. El Nahhas, Marta Ligero, Jakob N. Kather. Simultaneous prediction of tumor microenvironment biomarkers from pathology slides using multi-task deep regression [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6191.

Exploring the potential of deep regression model for next-location prediction

Exploring the potential of deep regression model for next-location prediction

Deep energy-pressure regression for a thermodynamically consistent EOS model

In this paper, we aim to explore novel machine learning (ML) techniques to facilitate and accelerate the construction of universal equation-Of-State (EOS) models with a high accuracy while ensuring important thermodynamic consistency. When applying ML to fit a universal EOS model, there are two key requirements: (1) a high prediction accuracy to ensure precise estimation of relevant physics properties and (2) physical interpretability to support important physics-related downstream applications. We first identify a set of fundamental challenges from the accuracy perspective, including an extremely wide range of input/output space and highly sparse training data. We demonstrate that while a neural network (NN) model may fit the EOS data well, the black-box nature makes it difficult to provide physically interpretable results, leading to weak accountability of prediction results outside the training range and lack of guarantee to meet important thermodynamic consistency constraints. To this end, we propose a principled deep regression model that can be trained following a meta-learning style to predict the desired quantities with a high accuracy using scarce training data. We further introduce a uniquely designed kernel-based regularizer for accurate uncertainty quantification. An ensemble technique is leveraged to battle model overfitting with improved prediction stability. Auto-differentiation is conducted to verify that necessary thermodynamic consistency conditions are maintained. Our evaluation results show an excellent fit of the EOS table and the predicted values are ready to use for important physics-related tasks.

Spectral encoder to extract the efficient features of Raman spectra for reliable and precise quantitative analysis

Raman spectroscopy has become a powerful analytical tool highly demanded in many applications such as microorganism sample analysis, food quality control, environmental science, and pharmaceutical analysis, owing to its non-invasiveness, simplicity, rapidity and ease of use. Among them, quantitative research using Raman spectroscopy is a crucial application field of spectral analysis. However, the entire process of quantitative modeling largely relies on the extraction of effective spectral features, particularly for measurements on complex samples or in environments with poor spectral signal quality. In this paper, we propose a method of utilizing a spectral encoder to extract effective spectral features, which can significantly enhance the reliability and precision of quantitative analysis. We built a latent encoded feature regression model; in the process of utilizing the autoencoder for reconstructing the spectrometer output, the latent feature obtained from the intermediate bottleneck layer is extracted. Then, these latent features are fed into a deep regression model for component concentration prediction. Through detailed ablation and comparative experiments, our proposed model demonstrates superior performance to common methods on single-component and multi-component mixture datasets, remarkably improving regression precision while without needing user-selected parameters and eliminating the interference of irrelevant and redundant information. Furthermore, in-depth analysis reveals that latent encoded feature possesses strong nonlinear feature representation capabilities, low computational costs, wide adaptability, and robustness against noise interference. This highlights its effectiveness in spectral regression tasks and indicates its potential in other application fields. Sufficient experimental results show that our proposed method provides a novel and effective feature extraction approach for spectral analysis, which is simple, suitable for various methods, and can meet the measurement needs of different real-world scenarios.

Spatial–Spectral–Temporal Deep Regression Model With Convolutional Long Short-Term Memory and Transformer for the Large-Area Mapping of Mangrove Canopy Height by Using Sentinel-1 and Sentinel-2 Data

Spatial–Spectral–Temporal Deep Regression Model With Convolutional Long Short-Term Memory and Transformer for the Large-Area Mapping of Mangrove Canopy Height by Using Sentinel-1 and Sentinel-2 Data

Microgrids and distribution system resilience assessment: A multi-objective robust-stochastic optimization approach considering social behavior using a deep learning modeling

In this paper, a new multi-objective two-stage robust-stochastic (MOTSRS) optimization approach for assessing microgrids and distribution system resilience is proposed. A novel approach based on deep learning is presented to model social behavior, specifically Social Reactions to Disasters (SRDs), in assessing microgrid and distribution system resilience. It is initially demonstrated that the proposed robust-stochastic approach is efficient for assessing the resilience of microgrids and distribution systems. A new resilience criterion is also proposed to be used for sensitivity analyses at the conservative level of the model. Subsequently, the effect of SRDs on system resilience is investigated through a supervised deep regression method. First, a large dataset is generated through a proposed nested Monte Carlo simulation. A numerical feature and five binary social behavior features are included in the dataset. Next, a feedforward neural network is utilized to predict the load during both normal and disaster circumstances based on the six features. The proposed supervised deep regression model with mean squared errors between 2e-5 and 2e-6 is capable of predicting electrical load during normal and disaster events. It is concluded that the resilience metrics of MGs, as well as expected operation costs and regret values, are influenced by the SRD effect.

Data‐Driven Design of Electrically Conductive Nanocomposite Materials: A Case Study of Acrylonitrile–Butadiene–Styrene/Carbon Nanotube Binary Composites

The field of polymer‐based nanoscience has always been of significant interest in the search for polymer/carbon nanotube (CNT) nanocomposites with optimized material properties for new applications. Herein, it is demonstrated that data collected from the online literature can be used to develop an efficient deep learning model to design acrylonitrile–butadiene–styrene (ABS)/CNT binary composites. A dataset of 14 945 data points is constructed from 110 studies. The results demonstrate that compared with a vanilla deep regression model, the proposed model achieves a 26% lower average mean absolute error and an accuracy of 80.6%, which is 16.2% higher than the vanilla deep regression model in predicting the six electrical and mechanical properties of target ABS/CNT composites. In addition, a Monte Carlo simulation integrated with the developed deep neural network (DNN) model effectively screens input variables for users and thus appropriately guides them to manufacture a composite product with desired physical properties.

Missing data reconstruction in attitude for quadrotor unmanned aerial vehicle based on deep regression model with different sensor failures

Forming a UAV cluster with multiple UAVs to collaborate in accomplishing tasks is the future development of battlefield. The UAV cluster can collaborate by communicating through inter-aircraft links and can quickly and accurately perform complex tasks such as path planning, collaborative reconnaissance, collaborative sensing and attack. In the process of Unmanned Aerial Vehicle (UAV) cooperative operation, obtaining the real-time attitude information of each UAV is preliminary to implement collaborative cooperation. However, in practice, due to the internal malfunction of sensors and mutual interference among UAVs, the acquired attitudes often suffer from data missing. This paper proposed a novel missing data reconstruction method for UAV’s attitude in the case of sensor failures. The attitude data of UAV is obtained through an advanced UAV simulation platform AirSim. Through fusing the temporal convolutional network (TCN) and Bi-directional long short-term memory (Bi-LSTM), a deep regression framework is established for missing data construction. In addition, we evaluate the reliability of the proposed method by comparing with different baseline models and different combinations of data missing. The performance of our method is quantified using metrics of root-mean-square-error (RMSE), mean absolute error (MAE), determination coefficient (DC), mean absolute percentage error (MAPE), and symmetric mean absolute percentage error (SMAPE) score. The averaged performance values of our proposed method outperform other baseline algorithms. The results of the experiment show that the study could well-address the issue of missing data reconstruction and provide reliable attitude data.

Recalling Unknown Manipulations by Spontaneously Sharing Actions with Similar Objects in Observation Based Learning

This paper proposes a method for a robot to recall multiple action candidates for an object by learning object manipulations based on observation of human actions. When learning, multiple answers to a single input in supervised regression manner, it is usually necessary to map all correct answers to the same input. However, only one action can be observed for an object at a time in observing object manipulations, and other possible actions are not always observed for the identical object. It is, therefore, important to automatically share various observed actions between similar-shaped objects by recognizing common shape cues among individual objects. The proposed method learns the code descriptions of object shapes by a variational auto-encoder (VAE) with an object image as input data, and the code descriptions of actions by a conditional VAE with object shape as a condition and an action as input data. Since the action is unknown recall target, it is desirable to obtain the code description of the action from only the object shape during recalling. The distribution of the code description of actions conditioned by input object shape on the obtained code description space is obtained by marginalization of the distribution learned by the encoder part of CVAE. However, since this marginalization is difficult to analytically and numerically operate, a deep regression model that “imitates” this marginal distribution is trained by using a maximum likelihood method based on sampling. Common actions of similar-shaped objects are shared among the similar objects in this “marginalization by imitation” process. Various possible actions for the input object shape can be recalled by repeatedly sampling from the imitated marginal distribution. This paper describes the results of experiment using actual object images and manipulation actions, and demonstrates the effectiveness of the proposed method.

DeepGA for automatically estimating fetal gestational age through ultrasound imaging

Accurate estimation of gestational age (GA) is vital for identifying fetal abnormalities. Conventionally, GA is estimated by measuring the morphology of the cranium, abdomen, and femur manually and inputting them into the classic Hadlock formula to assess fetal growth. However, this procedure incurs considerable overhead and suffers from bias caused by the operators, yielding suboptimal estimations. To address this challenge, we develop an automatic DeepGA model to achieve fully automatic GA prediction in an end-to-end manner. Our model uses a deep segmentation model (DeepSeg) to accurately identify and segment three critical tissues, including the cranium, abdomen, and femur, in which their morphology is automatically extracted. After that, we are able to directly estimate the GA via a deep regression model (DeepReg). We evaluate DeepGA on a large dataset, including 10,413 ultrasound images from 7113 subjects. It achieves superior performance over the traditional measurement approach, with a mean absolute estimation error (MAE) of 5 days. Our DeepGA model is a novel automatic solution on the basis of artificial intelligence learning that can help radiologists improve the performance of GA estimation in various clinical scenarios, thereby enhancing the efficiency of prenatal examinations.

Deep learning for fully-automated nuclear pleomorphism scoring in breast cancer

To guide the choice of treatment, every new breast cancer is assessed for aggressiveness (i.e., graded) by an experienced histopathologist. Typically, this tumor grade consists of three components, one of which is the nuclear pleomorphism score (the extent of abnormalities in the overall appearance of tumor nuclei). The degree of nuclear pleomorphism is subjectively classified from 1 to 3, where a score of 1 most closely resembles epithelial cells of normal breast epithelium and 3 shows the greatest abnormalities. Establishing numerical criteria for grading nuclear pleomorphism is challenging, and inter-observer agreement is poor. Therefore, we studied the use of deep learning to develop fully automated nuclear pleomorphism scoring in breast cancer. The reference standard used for training the algorithm consisted of the collective knowledge of an international panel of 10 pathologists on a curated set of regions of interest covering the entire spectrum of tumor morphology in breast cancer. To fully exploit the information provided by the pathologists, a first-of-its-kind deep regression model was trained to yield a continuous scoring rather than limiting the pleomorphism scoring to the standard three-tiered system. Our approach preserves the continuum of nuclear pleomorphism without necessitating a large data set with explicit annotations of tumor nuclei. Once translated to the traditional system, our approach achieves top pathologist-level performance in multiple experiments on regions of interest and whole-slide images, compared to a panel of 10 and 4 pathologists, respectively.

A document analysis deep learning regression model for initial coin offerings success prediction

Initial coin offerings (ICOs) provide an early-stage financing method for blockchain-based ventures. During the ICO process, whitepapers are important not only as promotional material through which ventures can demonstrate the technical and financial project details but also as references for investors. Persuasion theory and the related literature suggest that the presentation and order of information have a significant impact on the attitude of the audience. Therefore, in addition to projects’ metadata features, we construct a document analysis deep regression model (DADRM) to innovatively extract deep text and layout features from whitepapers. Based on a real-life dataset, we conduct a comparative study to assess the effectiveness of the proposed framework in predicting ICO success in terms of the funding amount. The empirical results show that our model that both extracts text content and retains the original 2D structure of the document can significantly reduce prediction error. Based on our proposed model, both ICO platforms and investors can prejudge the funding amount of cryptocurrency projects and mitigate information asymmetry. Additionally, this study demonstrates that both what is written in the business document and how the document is presented affect investor decisions.

Transformer-based highlights extraction from scientific papers

Highlights are short sentences used to annotate scientific papers. They complement the abstract content by conveying the main result findings. To automate the process of paper annotation, highlights extraction aims at extracting from 3 to 5 paper sentences via supervised learning. Existing approaches rely on ad hoc linguistic features, which depend on the analyzed context, and apply recurrent neural networks, which are not effective in learning long-range text dependencies.This paper leverages the attention mechanism adopted in transformer models to improve the accuracy of sentence relevance estimation. Unlike existing approaches, it relies on the end-to-end training of a deep regression model. To attend patterns relevant to highlights content it also enriches sentence encodings with a section-level contextualization. The experimental results, achieved on three different benchmark datasets, show that the designed architecture is able to achieve significant performance improvements compared to the state-of-the-art.

Abstract 4138: Prediction of progression-free survival in patients with primary glioblastoma: MRI T2 relaxivity and deep learning

Abstract Introduction: We aimed to evaluate T2 relaxation times in the follow-up of patients with glioblastoma (GBM), the most common and aggressive primary brain tumor with nearly universal recurrence following standard therapy. Methods: In this partially retrospective study, 53 patients (mean age 45.05 years ± 12.6) with non-recurrent GBM followed at our department roughly every 2 months after surgery and combined chemo-radiotherapy were included (>180 days progression-free survival [PFS] as determined by RANO,1 ≥3 MRI examinations in the PFS interval [same scanner and protocol]). Thirty-six patients additionally received Tumor-Treating Fields (TTFields) therapy. All data used were prior to progression. T2 relaxation rates (1/T2) were calculated voxel-wise assuming mono-exponential decay. Whole-brain (WB) histogram values were extracted from 1/T2 maps including skewness, kurtosis, mean, median, 15 and 85 percentile values and variance. We additionally segmented 1/T2 maps using a 5 compartment Gaussian mixture model (Python v3.8.2), producing mean, variance and voxel percentage for each component. To evaluate predictive potential with respect to PFS, we used a deep recurrent neural network (long short-term memory [LSTM] model) in Tensorflow v2.3.1 (4 timesteps); the features used included the above metrics as well as TTFields treatment status and scanner. Twenty subjects were included in the predictive model, as inclusion criteria were stricter (progression, ≤120 days between all MRIs [mean 43.6 days], ≥4 PFS MRIs). Two models were tested: a regression model (days to progression) and a classification model (±18 months PFS; models differed in output layer). Both models were run 10 times; mean results are presented. Results: We found WB median 1/T2 correlated with PFS, as values decreased prior to progression. WB median 1/T2 linear regression slopes also differed in progression versus pseudo-progression, as values were relatively stable in pseudo-progression. The deep LSTM regression model achieved an R2 of 97%. The deep LSTM classification model achieved a mean macro precision of 86%, recall 85% and F1 accuracy of 85% in predicting progression within 18 months. Conclusions: We found very intriguing results with WB median 1/T2 measurements in distinguishing progression from pseudo-progression, and in suggesting progression despite otherwise unremarkable imaging. A more complex predictive model trained on pre-progression data, using a fully-automated segmentation method followed by deep learning, showed very promising results. The present results may find utility in the monitoring of GBM patients and may be advantageous to include in multimodal predictive models. Reference: 1. Wen PY et al. J Clin Oncol 2010;28:1963-1972. Citation Format: Aaron M. Rulseh, Josef Vymazal. Prediction of progression-free survival in patients with primary glioblastoma: MRI T2 relaxivity and deep learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 4138.