Deep convolutional neural networks offer state-of-the-art performance for medical image analysis. However, their architectures are manually designed for particular problems. On the one hand, a manual designing process requires many trials to tune a large number of hyperparameters and is thus quite a time-consuming task. On the other hand, the fittest hyperparameters that can adapt to source data properties (e.g., sparsity, noisy features) are not able to be quickly identified for target data properties. For instance, the realistic noise in medical images is usually mixed and complicated, and sometimes unknown, leading to challenges in applying existing methods directly and creating effective denoising neural networks easily. In this paper, we present a Genetic Algorithm (GA)-based network evolution approach to search for the fittest genes to optimize network structures automatically. We expedite the evolutionary process through an experiencebased greedy exploration strategy and transfer learning. Our evolutionary algorithm procedure has flexibility, which allows taking advantage of current state-of-the-art modules (e.g., residual blocks) to search for promising neural networks. We evaluate our framework on a classic medical image analysis task: denosing. The experimental results on computed tomography perfusion (CTP) image denoising demonstrates the capability of the method to select the fittest genes for building highperformance networks, named EvoNets. Our results outperform state-of-the-art methods consistently at various noise levels.

With recent changes in the recommended annual limit on eye lens exposures to ionizing radiation, there is considerable interest in predictive computational dosimetry models of the human eye and its various ocular structures including the crystalline lens, ciliary body, cornea, retina, optic nerve, and central retinal artery. Computational eye models to date have been constructed as stylized models, high-resolution voxel models, and polygon mesh models. Their common feature, however, is that they are typically constructed of nominal size and of a roughly spherical shape associated with the emmetropic eye. In this study, we present a geometric eye model that is both scalable (allowing for changes in eye size) and deformable (allowing for changes in eye shape), and that is suitable for use in radiation transport studies of ocular exposures and radiation treatments of eye disease. The model allows continuous and variable changes in eye size (axial lengths from 20 to 26 mm) and eye shape (diopters from  -12 to  +6). As an explanatory example of its use, five models (emmetropic eyes of small, average, and large size, as well as average size eyes of  -12D and  +6D) were constructed and subjected to normally incident beams of monoenergetic electrons and photons, with resultant energy-dependent dose coefficients presented for both anterior and posterior eye structures. Electron dose coefficients were found to vary with changes to both eye size and shape for the posterior eye structures, while their values for the crystalline lens were found to be sensitive to changes in only eye size. No dependence upon eye size or eye shape was found for photon dose coefficients at energies below 2 MeV. Future applications of the model can include more extensive tabulations of dose coefficients to all ocular structures (not only the lens) as a function of eye size and shape, as well as the assessment of x-ray therapies for ocular disease for patients with non-emmetropic eyes.

The dosimetric dependence of ocular structures on eye size and shape was investigated within the standard ICRP Publication 116 irradiation geometries. A realistic transport geometry was constructed by inserting a scalable and deformable stylised eye model developed in our previous study within the head of the ICRP Publication 110 adult male reference computational phantom. Beam irradiations of external electrons, photons, and neutrons on this phantom were simulated using the Monte Carlo radiation transport code PHITS in the geometries of AP, RLAT, PA and ROT. Absorbed doses in ocular structures such as ciliary body, retina, and optic nerves were computed as well as that in lens. A clear dosimetric dependence of ocular structures on eye size and shape was observed for external electrons while only a small dependence was seen for external photons and neutrons. Difference of the tendency was attributed to their depth-dose distributions where spread dose distributions were created by photons and neutrons while more concentrated distributions were created by external electrons.