Lekan Ogunmolu


Publications Papers, arXiv submissions, etc

Videos and Talks Invited talks/videos

Open-Source Codes you might find useful.

On the web

Github

Blog

Email: olalekan DOT ogunmolu AT utsouthwestern DOT edu

Publications


PhD Thesis: A Multi-DOF Soft Robot Mechanism for Patient Motion Correction and Beam Orientation Selection in Cancer Radiation Therapy.

Accurate patient immobilization in conformal radiation therapy is crucial for efficient cancer treatment. Good treatment outcomes require accurate patient immobilization and a good choice of beam orientations. State-of-the-art immobilization systems rely on metallic or rigid masks which lack morphological properties, attenuate ionizing radiation, degrade dose efficacy, and are uncomfortable for the patient during treatment. The de-facto open-loop and deferred positioning procedures sometimes cause eczema or brain damage. We synthesize system identification, finite elastic deformation, and control systems to harness soft robot mechanisms for real-time motion correction in cancer radiation therapy scenarios.
Additionally, in most inverse treatment planning schemes today, the "right" beam angles among the myriad possibilities in beam space are usually determined through intuition and experience by treatment planners in a time-consuming trial-and-error procedure. Existing mathematical optimization techniques fail to meet a (near) real-time planning requirement. We propose a supervised pre-training of a deep neural network to assure quality beam plans are predicted in a real-time feasible manner. Our approach has the advantage of predicting feasible beam angles in near real time, and it is adaptable to treatment modalities that require large beam plans and $4\pi$-noncoplanar radiation therapy such as VMAT.

Olalekan Ogunmolu. circa May 2019..
PDF   Defense Slides   Code (Github)    

Using Supervised Learning and Guided Monte Carlo Tree Search for Beam Orientation Optimization in Radiation Therapy.

Azar Sadeghnejad Barkousaraie, Olalekan Ogunmolu, Steve Jiang, and Dan Nguyen. International Conference on Medical Image Computing and Computer Assisted Intervention, XXII (MICCAI), Shenzhen, China. October 2019.

A Fast Deep Learning Approach for Beam Orientation Selection Using Supervised Learning with Column Generation on IMRT Prostate Cancer Patients

We propose a fast beam orientation selection method, based on deep neural networks (DNN), capable of developing a plan comparable to those by the state-of-the-art column generation method. The novelty of Our model lies in its supervised learning structure, the DNN architecture, and ability to learn from anatomical features to predict dosimetrically suitable beam orientations without using the dosimetric information from the candidate beams, a time consuming and computationally expensive process. This may save hours of computation. A supervised DNN is trained to mimic the column generation algorithm, which iteratively chooses beam orientations by calculating beam fitness values based on the KKT optimality conditions. The dataset contains 70 prostate cancer patients. The DNN trained over 400 epochs, each with 2500 steps, using the Adam optimizer and a 6-fold cross-validation technique. The average and standard deviation of training, validation, and testing loss functions among the 6-folds were at most 1.44%. The differences in the dose coverage of PTV between plans generated by column generation and by DNN were 0.2%. The average dose differences received by organs at risk were between 1 and 6 percent: bladder had the smallest average difference, then rectum, left and right femoral heads. The dose received by body had an average difference of 0.1%. In the training phase of the proposed method, the model learns the suitable beam orientations based on the anatomical features of patients and omits time intensive calculations of dose influence matrices for all possible candidate beams. Solving the Fluence Map Optimization to get the final treatment plan requires calculating dose influence matrices only for the selected beams. The proposed DNN is a fast beam orientation selection method based that selects beam orientations in seconds and is therefore suitable for clinical routines.
Azar Sadeghnejad Barkousaraie, Olalekan Ogunmolu, Steve Jiang, and Dan Nguyen. Medical Physics. April 2019.
arXiv   Code (Github)    

An Approximate Policy Improvement Scheme for Beam Orientation Selection in Radiation Therapy

Olalekan Ogunmolu, Azar Sadeghnejad Barkousaraie, Dan Nguyen, and Steve Jiang. 61st Annual Meeting & Exhibition of the American Association of Physicists in Medicine (AAPM) Annual Meeting. San Antonio, TX, USA. July 2019.
PDF   Code (Github)    

A Monte Carlo Tree Game for Beam Orientation Optimization

We present an improvement to our formalism in WAFR wherein we devised the beam orientation problem in intensity-modulated radiation therapy as an approximate dynamic programming problem. Here, we improve the Monte-Carlo planning portion of the algorithm using a variant of approximate policy iteration: a batched alternating iteration of approximate policy evaluation and approximate policy improvement. Suppose that $\eta$ is the cardinality of all beam angles in a treatment setup, which together with a patient’s geometry, $\Upsilon$, constitute a deep neural network (DNN) policy input, $\Upsilon$=[$Y, \eta$]. A column generation DNN policy, $\Pi$, chooses a feasible angle based on a $\Upsilon$, and generates a profile map, $\zeta \in R^n$, whose argmax informs the next candidate beamlet that must selected (n=180 in our implementation) until we fulfill a beam plan requirement. This is the policy evaluation stage. We repeat the selection process for multiple patients and re-evaluate $\Pi$ based on a mean square error cost function. $\Pi$ essentially learns a mapping f:\Upsilon -->\zeta. With Monte-Carlo evaluations and tree search, we generate new training data in a policy improvement stage. We then project the new training data to the cost function space of $\Pi$. An alternating iteration of training $\Pi$ and improving $\Pi$ via our computed tree search policy, $\Gamma$, produces an improved policy, $\Pi^\star$ which we eventually use to evaluate treatment plans.
Olalekan Ogunmolu, Azar Sadeghnejad Barkousaraie, Dan Nguyen, and Steve Jiang.
The 19th International Conference on the use of Computers in Radiotherapy (ICCR), Montreal, Canada. PDF   Code (Github)    

A Monte Carlo Tree Game for Beam Orientation Optimization

We present an improvement to our formalism in WAFR wherein we devised the beam orientation problem in intensity-modulated radiation therapy as an approximate dynamic programming problem. Here, we improve the Monte-Carlo planning portion of the algorithm using a variant of approximate policy iteration: a batched alternating iteration of approximate policy evaluation and approximate policy improvement. Suppose that $\eta$ is the cardinality of all beam angles in a treatment setup, which together with a patient’s geometry, $\Upsilon$, constitute a deep neural network (DNN) policy input, $\Upsilon$=[$Y, \eta$]. A column generation DNN policy, $\Pi$, chooses a feasible angle based on a $\Upsilon$, and generates a profile map, $\zeta \in R^n$, whose argmax informs the next candidate beamlet that must selected (n=180 in our implementation) until we fulfill a beam plan requirement. This is the policy evaluation stage. We repeat the selection process for multiple patients and re-evaluate $\Pi$ based on a mean square error cost function. $\Pi$ essentially learns a mapping f:\Upsilon -->\zeta. With Monte-Carlo evaluations and tree search, we generate new training data in a policy improvement stage. We then project the new training data to the cost function space of $\Pi$. An alternating iteration of training $\Pi$ and improving $\Pi$ via our computed tree search policy, $\Gamma$, produces an improved policy, $\Pi^\star$ which we eventually use to evaluate treatment plans.
Olalekan Ogunmolu, Azar Sadeghnejad Barkousaraie, Dan Nguyen, and Steve Jiang. The 19th International Conference on the use of Computers in Radiotherapy (ICCR), Montreal, Canada.

PDF   Code (Github)    

Deep Learning Neural Network for Beam Orientation Optimization

In intensity modulated radiation therapy, the optimal choice of beam orientations has a significant impact on the treatment plan quality, influencing the final treatment outcome. In current treatment planning workflow, the beam direction selection is manually done by the planner, typically yielding suboptimal solutions. Beam Orientation Optimization (BOO) methods are used to find the optimal beam directions. Conventional BOO methods need to previously compute the dose influence matrices (Dij) for multiple beam orientations (around 100-200 to cover all potential beam angles), and use them to perform Fluence Map Optimization (FMO). However, both the computation of Dij matrices and FMO are very complex and time intensive operations (can be several hours for all BOO candidate beams), which hampers their implementations in clinical routine. In contrast, artificial intelligence (AI) is an attractive tool for solving BOO problems, given its superior speed and promising results for medical applications. This work aims to develop a fast BOO method based on deep neural networks that provides a solution in about a second, and therefore, can be implemented in clinical routine.
Azar Sadeghnejad Barkousaraie, Olalekan Ogunmolu, Steve Jiang, and Dan Nguyen.
Accepted at The 19th International Conference on the use of Computers in Radiotherapy (ICCR), Montreal, Canada.
PDF   Code (Github)    

Deep BOO! Automating Beam Orientation Optimization in Intensity Modulated Radiation Therapy.

Intensity-Modulated Radiation Therapy (IMRT) is a method for treating cancers by aiming radiation to cancer tumor(s) while minimizing radiation to organs-at-risk from a robot's tool frame. Computationally finding the correct treatment plan for a target volume is often an exhaustive combinatorial search problem where traditional optimization methods have not yielded real-time feasible results. Aiming to automate the beam orientation and intensity modulation process, we introduce a novel set of techniques leveraging (i) pattern recognition, (ii) monte carlo evaluations, (iii) game theory, and (iv) neuro-dynamic programming. We optimize a deep neural network policy that guides Monte Carlo simulations of promising beamlets. Seeking a saddle equilibrium, we let two fictitious neural network players, within a zero-sum markov game framework, alternatingly play a best response to their opponent's mixed strategy profile during episodes of a two-player markov decision game. During inference, the optimized policy predicts feasible beam angles on test target volumes. This work merges the beam orientation and fluence map optimization subproblems in IMRT sequential treatment planning system into one pipeline. We formally introduce our approach, and present numerical results for coplanar beam angles on prostate cases.
Olalekan Ogunmolu, Michael Folkerts, Dan Nguyen, Nicholas Gans and Steve Jiang.
Algorithm Foundations of Robotics Workshop (WAFR) XIII, Merida, Mexico.
Published in Springer’s Proceedings in Advanced Robotics (SPAR) Book (2019).
PDF      Code (Github)     WAFR Submission PDF      Camera Ready PDF  

Automating Beam Orientation Optimization during IMRT Cancer Treatment Planning: A Deep Reinforcement Learning Approach.

This work focuses on automating beam orientation optimization (BOO) during IMRT treatment planning. Being a nonconvex problem, we develop a hybrid approach – combining techniques from deep reinforcement learning (DRL), and fluence map optimization (FMO), to find beam orientations that yield satisfactory local minima, in terms of IMRT treatment plan quality.
Olalekan Ogunmolu*, Dan Nguyen, Chenyang Shen, Xun Jia, Weiguo Lu, Nicholas Gans, and Steve Jiang.
American Association of Physicists in Medicine (AAPM) Meeting . Nashville, TN. July 2018.
PDF   AAPM Program  

Minimax Iterative Dynamic Game : Application to Nonlinear Robot Control Tasks.

Multistage decision policies provide useful control strategies in high-dimensional state spaces, particularly in complex control tasks. However, they exhibit weak performance guarantees in the presence of disturbance, model mismatch, or model uncertainties. This brittleness limits their use in high-risk scenarios. We present how to quantify the sensitivity of such policies in order to inform of their robustness capacity. We also propose a minimax iterative dynamic game framework for designing robust policies in the presence of disturbance/uncertainties. We test the quantification hypothesis on a carefully designed deep neural network policy; we then pose a minimax iterative dynamic game (iDG) framework for improving policy robustness in the presence of adversarial disturbances. We evaluate our iDG framework on a mecanum-wheeled robot, whose goal is to find a ocally robust optimal multistage policy that achieve a given goal-reaching task. The algorithm is simple and adaptable for designing meta-learning/deep policies that are robust against disturbances, model mismatch, or model uncertainties, up to a disturbance bound. Videos of the results are on the author's website, while the codes for reproducing our experiments are on github. A self-contained environment for reproducing our results is on docker.
Olalekan Ogunmolu*, Nicholas Gans, and Tyler Summers.
Intelligent Robots and Systems (IROS) 2018 .Madrid, Spain. 2018.
PDF   Code (Github)  Code (Docker Image)  Videos website 

Minimax Iterative Dynamic Game : Application to Nonlinear Robot Control Tasks.

Multistage decision policies provide useful control strategies in high-dimensional state spaces, particularly in complex control tasks. However, they exhibit weak performance guarantees in the presence of disturbance, model mismatch, or model uncertainties. This brittleness limits their use in high-risk scenarios. We present how to quantify the sensitivity of such policies in order to inform of their robustness capacity. We also propose a minimax iterative dynamic game framework for designing robust policies in the presence of disturbance/uncertainties. We test the quantification hypothesis on a carefully designed deep neural network policy; we then pose a minimax iterative dynamic game (iDG) framework for improving policy robustness in the presence of adversarial disturbances. We evaluate our iDG framework on a mecanum-wheeled robot, whose goal is to find a ocally robust optimal multistage policy that achieve a given goal-reaching task. The algorithm is simple and adaptable for designing meta-learning/deep policies that are robust against disturbances, model mismatch, or model uncertainties, up to a disturbance bound. Videos of the results are on the author's website, while the codes for reproducing our experiments are on github. A self-contained environment for reproducing our results is on docker.
Olalekan Ogunmolu*, Nicholas Gans, and Tyler Summers.
Machine Learning for Planning and Control Workshop, International Conference on Robotics and Automation 2018 .Brisbane, Australia. 2018.
PDF   Code (Github)  Code (Docker Image)  Videos website 

Design and Development of Soft Robots for Head and Neck Cancer Radiotherapy.

Yara Almubarak, Olalekan Ogunmolu*, Xuejun Gu, Steve Jiang, Nicholas Gans, and Yonas Tadesse.
SPIE: Smart Structures + Nondestructive Evaluation. Denver, CO, U.S.A. March 2018.

Soft-NeuroAdapt: A 3-DOF neuro-adaptive patient pose correction system for frameless and maskless cancer radiotherapy.

Precise patient positioning is fundamental to successful removal of malignant tumors during treatment of head and neck cancers. Errors in patient positioning have been known to damage critical organs and cause complications. To better address issues of patient positioning and motion, we introduce a 3-DOF neuro-adaptive soft-robot, called Soft-NeuroAdapt to correct deviations along 3 axes. The robot consists of inflatable air bladders that adaptively control head deviations from target while ensuring patient safety and comfort. The adaptive-neuro controller combines a state feedback component, a feedforward regulator, and a neural network that ensures correct adaptation. States are measured by a 3D vision system. We validate Soft-NeuroAdapt on a 3D printed head-and-neck dummy, and demonstrate that the controller provides adaptive actuation that compensates for intrafractional deviations in patient positioning.
Olalekan Ogunmolu*, Adwait Kulkarni, Yonas Tadesse, Xuejun Gu, Steve Jiang, and Nicholas Gans.
In proceedings of Intelligent Robots and Systems (IROS) 2017. Vancouver, BC, Canada. 2017.
PDF   arXiv   Code (Github) 

Nonlinear Systems Identification Using Deep Dynamic Neural Networks.

An investigation of the effectiveness of deep neural networks in the modeling of dynamical systems with complex behavior. Three deep neural network structures are trained on sequential data, and we investigate the effectiveness of these networks in modeling associated characteristics of the underlying dynamical systems.
PDF   arXiv   Code (Github)  Blog  


Vision-based control of a soft-robot for Maskless Cancer Radiotherapy.

An LQG-based controller for a soft-robot targeted towards accurate patient positioning systems in maskless head and neck cancer radiotherapy. With two RGB-D sensors in a multisensor kalman fusion scenario, we estimate the position of a patient during simulated maskless cancer RT in real-time.
Olalekan Ogunmolu*, Xuejun Gu, Steve Jiang, and Nicholas Gans. IEEE Conference on Automation Science and Engineering (CASE). Fort-Worth, Texas, August 2016.
PDF   arXiv   Code (Github)   Slides


A Real-Time Soft-Robotic Patient Positioning System for Maskless Head-and-Neck Cancer Radiotherapy.

Use of inflatable air bladders in controlling the flexion/extension cranial motion of a simulated patient during maskless cancer radiotherapy. We imitate IGRT by using RGB-D depth sensors to generate surface images of a patient and extract the pose information from the reconstructed mesh. Results show that the system is capable of controlling head motion to within 2mm with respect to a reference trajectory.
Olalekan Ogunmolu*, Xuejun Gu, Steve Jiang, and Nicholas Gans. IEEE Conference on Automation Science and Engineering (CASE). Gothenburg, Sweden, August 2015.
PDF   arXiv   Code (Github)   Slides  doi: 10.1109/CoASE.2015.7294318


An Image-Guided Soft Robotic Patient Positioning System for Maskless Head-And-Neck Cancer Radiotherapy: A Proof-Of-Concept Study.

Use of inflatable air bladders in controlling the flexion/extension cranial motion of a simulated patient during maskless cancer radiotherapy. We imitate IGRT by using RGB-D depth sensors to generate surface images of a patient and extract the pose information from the reconstructed mesh. Results show that the system is capable of controlling head motion to within 2mm with respect to a reference trajectory.
Olalekan Ogunmolu*, Nick Gans, Steve Jiang, Xuejun Gu. American Association of Physicists in Medicine (AAPM) Annual Meeting. Anaheim, CA, July 2015.
Abstract  


Autonomous Navigation of a Rotor-craft unmanned aerial vehicle using machine vision.

Flying of the ACSE dept quadrotor using extracted image features from a structured indoor environment. We apply invariant moments, extract corners from UAV camera frames and compare results with calibrated helipad image; we learn actual helipad features by computing matching variance and then estimate the UAV pose in a hovering condition to plan landing.
Olalekan Ogunmolu.. MS Thesis. ACSE Dept, The University of Sheffield. Sheffield, United Kingdom, August 2012.
PDF  


Curriculum Vitae

Academic CV (.pdf).

Last updated: November, 2018
.



Concise Resume

Resume (.pdf).

Last updated: November, 2018
.