Convolutional neural networks (CNNs) have shown promising performance in various 2D computer vision tasks due to availability of large amounts of 2D training data. Contrarily, medical imaging deals with 3D data and usually lacks the equivalent extent and diversity of data, for developing AI models. Transfer learning provides the means to use models trained for one application as a starting point to another application. In this work, we leverage 2D pre-trained models as a starting point in 3D medical applications by exploring the concept of Axial-Coronal-Sagittal (ACS) convolutions. We have incorporated ACS as an alternative of native 3D convolutions in the Generally Nuanced Deep Learning Framework (GaNDLF), providing various well-established and state-of-the-art network architectures with the availability of pre-trained encoders from 2D data. Results of our experimental evaluation on 3D MRI data of brain tumor patients for i) tumor segmentation and ii) radiogenomic classification, show model size reduction by ~22% and improvement in validation accuracy by ~33%. Our findings support the advantage of ACS convolutions in pre-trained 2D CNNs over 3D CNN without pre-training, for 3D segmentation and classification tasks, democratizing existing models trained in datasets of unprecedented size and showing promise in the field of healthcare.
Brain extraction is an indispensable step in neuro-imaging with a direct impact on downstream analyses. Most such methods have been developed for non-pathologically affected brains, and hence tend to suffer in performance when applied on brains with pathologies, e.g., gliomas, multiple sclerosis, traumatic brain injuries. Deep Learning (DL) methodologies for healthcare have shown promising results, but their clinical translation has been limited, primarily due to these methods suffering from i) high computational cost, and ii) specific hardware requirements, e.g., DL acceleration cards. In this study, we explore the potential of mathematical optimizations, towards making DL methods amenable to application in low resource environments. We focus on both the qualitative and quantitative evaluation of such optimizations on an existing DL brain extraction method, designed for pathologically-affected brains and agnostic to the input modality. We conduct direct optimizations and quantization of the trained model (i.e., prior to inference on new data). Our results yield substantial gains, in terms of speedup, latency, throughput, and reduction in memory usage, while the segmentation performance of the initial and the optimized models remains stable, i.e., as quantified by both the Dice Similarity Coefficient and the Hausdorff Distance. These findings support post-training optimizations as a promising approach for enabling the execution of advanced DL methodologies on plain commercial-grade CPUs, and hence contributing to their translation in limited- and low- resource clinical environments.
Convolutional neural networks (CNNs) have achieved remarkable success in automatically segmenting organs or lesions on 3D medical images. Recently, vision transformer networks have exhibited exceptional performance in 2D image classification tasks. Compared with CNNs, transformer networks have an appealing advantage of extracting long-range features due to their self-attention algorithm. Therefore, we propose a CNN-Transformer combined model, called BiTr-Unet, with specific modifications for brain tumor segmentation on multi-modal MRI scans. Our BiTr-Unet achieves good performance on the BraTS2021 validation dataset with median Dice score 0.9335, 0.9304 and 0.8899, and median Hausdorff distance 2.8284, 2.2361 and 1.4142 for the whole tumor, tumor core, and enhancing tumor, respectively. On the BraTS2021 testing dataset, the corresponding results are 0.9257, 0.9350 and 0.8874 for Dice score, and 3, 2.2361 and 1.4142 for Hausdorff distance. The code is publicly available at https://github.com/JustaTinyDot/BiTr-Unet .
Deformable registration of magnetic resonance images between patients with brain tumors and healthy subjects has been an important tool to specify tumor geometry through location alignment and facilitate pathological analysis. Since tumor region does not match with any ordinary brain tissue, it has been difficult to deformably register a patients brain to a normal one. Many patient images are associated with irregularly distributed lesions, resulting in further distortion of normal tissue structures and complicating registration's similarity measure. In this work, we follow a multi-step context-aware image inpainting framework to generate synthetic tissue intensities in the tumor region. The coarse image-to-image translation is applied to make a rough inference of the missing parts. Then, a feature-level patch-match refinement module is applied to refine the details by modeling the semantic relevance between patch-wise features. A symmetry constraint reflecting a large degree of anatomical symmetry in the brain is further proposed to achieve better structure understanding. Deformable registration is applied between inpainted patient images and normal brains, and the resulting deformation field is eventually used to deform original patient data for the final alignment. The method was applied to the Multimodal Brain Tumor Segmentation (BraTS) 2018 challenge database and compared against three existing inpainting methods. The proposed method yielded results with increased peak signal-to-noise ratio, structural similarity index, inception score, and reduced L1 error, leading to successful patient-to-normal brain image registration.
Glioblastoma ( GBM ) is arguably the most aggressive, infiltrative, and heterogeneous type of adult brain tumor. Biophysical modeling of GBM growth has contributed to more informed clinical decision-making. However, deploying a biophysical model to a clinical environment is challenging since underlying computations are quite expensive and can take several hours using existing technologies. Here we present a scheme to accelerate the computation. In particular, we present a deep learning ( DL )-based logistic regression model to estimate the GBM's biophysical growth in seconds. This growth is defined by three tumor-specific parameters: 1) a diffusion coefficient in white matter ( Dw ), which prescribes the rate of infiltration of tumor cells in white matter, 2) a mass-effect parameter ( Mp ), which defines the average tumor expansion, and 3) the estimated time ( T ) in number of days that the tumor has been growing. Preoperative structural multi-parametric MRI ( mpMRI ) scans from n = 135 subjects of the TCGA-GBM imaging collection are used to quantitatively evaluate our approach. We consider the mpMRI intensities within the region defined by the abnormal FLAIR signal envelope for training one DL model for each of the tumor-specific growth parameters. We train and validate the DL-based predictions against parameters derived from biophysical inversion models. The average Pearson correlation coefficients between our DL-based estimations and the biophysical parameters are 0.85 for Dw, 0.90 for Mp, and 0.94 for T, respectively. This study unlocks the power of tumor-specific parameters from biophysical tumor growth estimation. It paves the way towards their clinical translation and opens the door for leveraging advanced radiomic descriptors in future studies by means of a significantly faster parameter reconstruction compared to biophysical growth modeling approaches.
