I. El Naqa

Computer-Aided Detection and Diagnosis of Breast Cancer With Mammography: Recent Advances

Breast cancer is the second-most common and leading cause of cancer death among women. It has become a major health issue in the world over the past 50 years, and its incidence has increased in recent years. Early detection is an effective way to diagnose and manage breast cancer. Computer-aided detection or diagnosis (CAD) systems can play a key role in the early detection of breast cancer and can reduce the death rate among women with breast cancer. The purpose of this paper is to provide an overview of recent advances in the development of CAD systems and related techniques. We begin with a brief introduction to some basic concepts related to breast cancer detection and diagnosis. We then focus on key CAD techniques developed recently for breast cancer, including detection of calcifications, detection of masses, detection of architectural distortion, detection of bilateral asymmetry, image enhancement, and image retrieval.

Exploring feature-based approaches in PET images for predicting cancer treatment outcomes

Accumulating evidence suggests that characteristics of pre-treatment FDG-PET could be used as prognostic factors to predict outcomes in different cancer sites. Current risk analyses are limited to visual assessment or direct uptake value measurements. We are investigating intensity-volume histogram metrics and shape and texture features extracted from PET images to predict patients response to treatment. These approaches were demonstrated using datasets from cervix and head and neck cancers, where AUC of 0.76 and 1.0 were achieved, respectively. The preliminary results suggest that the proposed approaches could potentially provide better tools and discriminant power for utilizing functional imaging in clinical prognosis.

Dose response explorer: an integrated open-source tool for exploring and modelling radiotherapy dose–volume outcome relationships

Radiotherapy treatment outcome models are a complicated function of treatment, clinical and biological factors. Our objective is to provide clinicians and scientists with an accurate, flexible and user-friendly software tool to explore radiotherapy outcomes data and build statistical tumour control or normal tissue complications models. The software tool, called the dose response explorer system (DREES), is based on Matlab, and uses a named-field structure array data type. DREES/Matlab in combination with another open-source tool (CERR) provides an environment for analysing treatment outcomes. DREES provides many radiotherapy outcome modelling features, including (1) fitting of analytical normal tissue complication probability (NTCP) and tumour control probability (TCP) models, (2) combined modelling of multiple dose-volume variables (e.g., mean dose, max dose, etc) and clinical factors (age, gender, stage, etc) using multi-term regression modelling, (3) manual or automated selection of logistic or actuarial model variables using bootstrap statistical resampling, (4) estimation of uncertainty in model parameters, (5) performance assessment of univariate and multivariate analyses using Spearmans rank correlation and chi-square statistics, boxplots, nomograms, Kaplan-Meier survival plots, and receiver operating characteristics curves, and (6) graphical capabilities to visualize NTCP or TCP prediction versus selected variable models using various plots. DREES provides clinical researchers with a tool customized for radiotherapy outcome modelling. DREES is freely distributed. We expect to continue developing DREES based on user feedback.

Learning a Channelized Observer for Image Quality Assessment

It is now widely accepted that image quality should be evaluated using task-based criteria, such as human-observer performance in a lesion-detection task. The channelized Hotelling observer (CHO) has been widely used as a surrogate for human observers in evaluating lesion detectability. In this paper, we propose that the problem of developing a numerical observer can be viewed as a system-identification or supervised-learning problem, in which the goal is to identify the unknown system of the human observer. Following this approach, we explore the possibility of replacing the Hotelling detector within the CHO with an algorithm that learns the relationship between measured channel features and human observer scores. Specifically, we develop a channelized support vector machine (CSVM) which we compare to the CHO in terms of its ability to predict human-observer performance. In the examples studied, we find that the CSVM is better able to generalize to unseen images than the CHO, and therefore may represent a useful improvement on the CHO methodology, while retaining its essential features.

Nonlinear Kernel-Based Approaches for Predicting Normal Tissue Toxicities

Since the early demonstration of the curative potential of radiation therapy for tumor sterilization, normal tissue toxicity continues to be dose limiting. Accurate prediction of patient¿s complication risk would allow personalization of treatment planning decisions. Nonlinear kernel methods can provide a robust framework for learning complex interactions between observed toxicities and treatment, anatomical, and patient-related variables. However, proper application of these powerful methods would require better understanding of a high-dimensional feature space that is spanned by all these variables. In this work, we investigate methods for visualization of this high-dimensional space and compare different approaches for extracting discriminant features. Our preliminary results demonstrate that principle component analysis is a valuable tool for visualizing high dimensional data and for determining proper kernel type. In addition, variable selection based on resampling methods within the logistic regression framework seemed to yield improved prediction performance compared to the recursive-feature elimination method.

Automated breathing motion tracking for 4D computed tomography

4D-CT is being developed to provide breathing motion information for radiation therapy treatment planning. Potential applications include optimization of intensity-modulated beams in the presence of breathing motion and intra-fraction target volume margin determination for conformal therapy. A major challenge of this process is the determination of the internal motion (trajectories) from the 4D CT data. Manual identification and tracking of internal landmarks is impractical. For example, in a single couch position, 512 /spl times/ 512 /spl times/ 12 pixel CT scans contains 3.1/spl times/10/sup 5/ voxels. If 15 of these scans are acquired throughout the breathing cycle, there are almost 47 million voxels to evaluate necessitating automation of the registration process. The natural high contrast between bronchi, vessels, other lung tissue offers an excellent opportunity to develop automated deformable registration techniques. We have been investigating the use motion compensated temporal smoothing using optical flow for this purpose. Optical flow analysis uses the CT intensity and temporal (in our case tidal volume) gradients to estimate the motion trajectories. The algorithm is applied to 3D image datasets reconstructed at different percentiles of tidal volumes. The trajectories can be used to interpolate CT datasets between tidal volumes.

Compensation of breathing motion artifacts in thoracic PET images by wavelet-based deconvolution

In biological imaging of thoracic tumors using FDG-PET, blurring due to breathing motion often significantly degrades the quality of the observed image, which then obscures the tumor boundary. The effect could be detrimental in small lesions. We demonstrate a deconvolution technique that combines patient-specific motion estimates of tissue trajectories with wavelet decomposition to compensate for breathing-motion induced artifacts. The lung motion estimates were obtained using a breathing model that maps spatial trajectories in CT data as a function of tidal volume and airflow measured by spirometry. Initial results showed good improvement in the spatial resolution, especially in the direction of major lung motion (craniocaudal) on phantom data as well as on clinical data with large or small tumors

TU‐D‐204C‐04: Machine Learning as New Tool for Predicting Radiotherapy Response

Radiotherapy outcomes are determined by complex interactions between treatment techniques, cancerpathology, and patient‐related physiological and biological factors. A common obstacle to building maximally predictive treatment outcome models for clinical practice in radiation oncology is the failure to capture this complexity of heterogeneous variable interactions and the ability to apply outcome models across different multi‐institutional data. Methods based on machine learning can identify data patterns, variable interactions, and higher order relationships among prognostic variables. In addition, they have the ability to generalize to unseen data before. In this work, we will provide an overview of the current role of machine learning methods for predicting post‐radiotherapy tumor control probability (TCP) and normal tissue toxicities (NTCP). We will discuss some of the current challenges in the field and highlight the potential opportunities of machine learning methods for future treatment outcomes research in radiation oncology.

A fast inverse consistent deformable image registration method based on symmetric optical flow computation

Deformable image registration is widely used in various radiation therapy applications including 4D-CT and treatment planning adaptation. In this work, a simple and efficient inverse consistency deformable registration method is proposed with aims of higher registration accuracy and faster convergence speed. Instead of registering image I to the second image J, two images are symmetrically deformed toward one another in multiple passes, until both deformed images are registered. In every pass, a delta motion field is computed by minimizing a symmetric optical flow system cost function using the modified optical flow algorithms. The images are then further deformed with the delta motion field in positive and negative directions, respectively, and then used for the next pass. The magnitude of the delta motion field is forced to be less than 0.4 voxel for every pass in order to guarantee the smoothness and invertibility of the two overall motion fields which are accumulating the delta motion fields in positive and negative directions, respectively. The final motion fields to register the original images I and J, in either direction, are calculated by inverting one overall motion field and composing the inversion result with the other overall motion field. The final motion fields are inversely consistent and this is ensured by the symmetric way that registration is carried out. Results suggest that the method is able to improve the overall accuracy by 30% or more, reduce the inverse consistency error, and increase the convergence rate. The computation speed may slightly decrease, or increase in some cases because the new method converges faster. Comparing to previously published inverse consistency algorithms, the proposed method is simpler in theory, easier to implement, and faster.

SU-FF-T-375: Machine Learning Methods for Radiobiological Outcomes Modeling

Purpose: Radiobiological outcomes models are important predictors of irradiation induced effects in terms of achieving tumor control or causing damage to surrounding normal tissues. They are also used to rank the quality of treatment plans. Outcomes models may depend on many variables such as dose-volume metrics and clinical factors. In particular, the best outcome model itself may vary depending on patient or treatment characteristics. General non-linear models, such as neural-networks, potentially allow us to capture this natural variation in models. Method and Materials: We studied feed-forward (FFNN) and general regression neural networks (GRNN). As representative data, we used a cohort 166 non-small cell lung cancer patients who received radiotherapy treatment, with endpoints of pneumonitis and esophagitis. Dosimetric variables were extracted using CERR. Results: We used resampling (bootstrap) methods to select optimal parameters for the networks, which include the number of neurons in FFNNs and the ‘width’ (σ) in GRNNs. In modeling pneumonitis, the optimal FFNN had 3 layers and 5 neurons in the hidden layer, with spearman rank correlation 0.49±0.27 in training and 0.11±0.07 in testing. The GRNN with σ=1.25, achieved a training spearman of 0.25±0.08 and testing spearman of 0.20±0.3. In modeling esophagitis, the FFNN had 5 neurons, with a spearman of 0.59±0.09 in training and 0.3±0.21 in testing. GRNN with σ=1.25, achieved a training spearman of 0.38±0.06 and a testing spearman of 0.39±0.12. Conclusion: We evaluated two machine learning algorithms to model outcome in cases of pneumonitis and esophagitis. Our preliminary results indicate that the GRNN was more straightforward to implement and, more importantly, had better generalizability than that of FFNN. Our experience to date indicates that neural networks may perform as well or better than multi-term logistic regression methods.