Guest Editorial: Knowledge-based deep learning system in bio-medicine

Abstract

Numerous healthcare procedures can be viewed as medical sector decisions. In the modern era, computers have become indispensable in the realm of medical decision-making. However, the common view of computers in the medical field typically extends only to applications that support doctors in diagnosing diseases. To more tightly intertwine computers with the biomedical sciences, professionals are now more frequently utilising knowledge-driven deep learning systems (KDLS) and their foundational technologies, especially in the domain of neuroimaging (NI).

Data for medical purposes can be sourced from a variety of imaging techniques, including but not limited to Computed Tomography (CT), Magnetic Resonance Imaging (MRI), Ultrasound, Single Photon Emission Computed Tomography (SPECT), Positron Emission Tomography (PET), Magnetic Particle Imaging (MPI), Electroencephalography (EEG), Magnetoencephalography (MEG), Optical Microscopy and Tomography, Photoacoustic Tomography, Electron Tomography, and Atomic Force Microscopy.

Historically, these imaging techniques have been analysed using traditional statistical methods, such as hypothesis testing or Bayesian inference, which often presuppose certain conditions that are not always met. An emerging solution is the implementation of machine learning (ML) within the context of KDLS, allowing for the empirical mapping of complex, multi-dimensional relationships within data sets.

The objective of this special issue is to showcase the latest advancements in the methodology of KDLS for evaluating functional connectivity, neurological disorders, and clinical neuroscience, such as conditions like Alzheimer's, Parkinson's, cerebrovascular accidents, brain tumours, epilepsy, multiple sclerosis, ALS, Autism Spectrum Disorder, and more. Additionally, the special issue seeks to elucidate the mechanisms behind the predictive capabilities of ML methods within KDLS for brain-related diseases and disorders.

We received an abundance of submissions, totalling more than 40, from over 10 countries. After a meticulous and rigorous peer review process, which employed a double-blind methodology, we ultimately selected eight outstanding papers for publication. This process ensured the highest standards of quality and impartiality in the selection.

In the article ‘A deep learning fusion model for accurate classification of brain tumours in Magnetic Resonance images’, Zebari et al. created a robust deep learning (DL) fusion model for accurate brain tumour classification. To enhance performance, they employed data augmentation to expand the training dataset. The model leveraged VGG16, ResNet50, and convolutional deep belief networks to extract features from MRI images using a softmax classifier. By fusing features from two DL models, the fusion model notably boosted classification precision. Tested with a publicly available dataset, it achieved a remarkable 98.98% accuracy rate, outperforming existing methods.

In the article ‘Knowledge-based deep learning system for classifying’, Dhaygude et al. proposed a deep three-dimensional convolutional neural network incorporating multi-task learning and attention mechanisms. An upgraded primary C3D network is utilised to create rougher low-level feature maps. It introduces a new convolution block that focuses on the structural aspects of the magnetic resonance imaging image and another block that extracts attention weights unique to certain pixel positions in the feature map and multiplies them with the feature map output. Then, several fully connected layers are used to achieve multi-task learning, generating three outputs, including the primary classification task. The other two outputs employ backpropagation during training to improve the primary classification job. Experimental findings show that the authors' proposed method outperforms current approaches for classifying AD, achieving enhanced classification accuracy and other indicators on the Alzheimer's disease neuroimaging initiative dataset. The authors demonstrated promise for future disease classification studies.

In the paper entitled ‘A novel medical image data protection scheme for smart healthcare system’, Rehman et al. presented a lightweight medical image encryption scheme utilising bit-plane decomposition and chaos theory. The results of the experiment yielded entropy, energy, and correlation values of 7.999, 0.0156, and 0.0001, respectively. This validated the effectiveness of the encryption system proposed in this paper, which offered high-quality encryption, a large key space, key sensitivity, and resistance to statistical attacks.

In the paper entitled ‘Image super-resolution via dynamic network’, Tian et al. presented a dynamic network for image super-resolution (DSRNet), which contained a residual enhancement block, wide enhancement block, feature refinement block, and construction block. The residual enhancement block was composed of a residual enhanced architecture to facilitate hierarchical features for image super-resolution. To enhance the robustness of the obtained super-resolution model for complex scenes, a wide enhancement block achieved a dynamic architecture to learn more robust information to enhance the applicability of an obtained super-resolution model for varying scenes. To prevent interference of components in a wide enhancement block, a refinement block utilised a stacked architecture to learn obtained features accurately. Also, a residual learning operation was embedded in the refinement block to prevent long-term dependency problems. Finally, a construction block was responsible for reconstructing high-quality images. Designed heterogeneous architecture could not only facilitate richer structural information but also be lightweight, which was suitable for mobile digital devices. Experimental results showed that our method is more competitive in performance, recovering time of image super-resolution, and complexity. The code of DSRNet can be obtained at https://github.com/hellloxiaotian/DSRNet.

In the paper ‘Improved organs at risk segmentation based on modified U-Net with self-attention and consistency regularisation’, Manko et al. presented a new approach to automatic OAR segmentation in the chest cavity in Computed Tomography (CT) images. The proposed approach is based on the modified U-Net architecture with the ResNe-34 encoder, the baseline adopted in this work. The new two-branch CS-SA U-Net architecture is proposed, which consists of two parallel U-Net models in which self-attention blocks with cosine similarity as query-key similarity function (CS-SA) blocks are inserted between the encoder and decoder, which enabled the use of consistency regularisation. The proposed solution demonstrates state-of-the-art performance for the problem of OAR segmentation in CT images on the publicly available SegTHOR benchmark dataset in terms of a Dice coefficient (oesophagus-0.8714, heart-0.9516, trachea-0.9286, and aorta-0.9510) and Hausdorff distance (oesophagus-0.2541, heart-0.1514, trachea-0.1722, and aorta-0.1114) and significantly outperforms the baseline. The current approach was demonstrated to be viable for improving the quality of OAR segmentation for radiotherapy planning.

In the paper ‘GAN-MD: A myocarditis detection using multi-channel’, Golilarz et al. suggested a novel technique for improving the classification precision of CNNs. Using generative adversarial networks (GANs) to create synthetic images for data augmentation, the authors addressed challenges such as mode collapse and unstable training. Incorporating a reconstruction loss into the GAN loss function requires the generator to produce images reflecting the discriminator features, thus enhancing the generated images' quality to more accurately replicate authentic data patterns. Moreover, combining this loss function with other regularisation methods, such as gradient penalty, has proven to improve the performance of diverse GAN models further. A significant challenge in myocarditis diagnosis was the imbalance of classification, where one class dominated the other. To mitigate this, the authors introduced a focal loss-based training method that effectively trains the model on the minority class samples. The GAN-MD approach, evaluated on the Z-Alizadeh Sani myocarditis dataset, achieved superior results (F-measure 86.2%; geometric mean 91.0%) compared with other deep learning models and traditional machine learning methods.

In the paper entitled ‘DeepGCN based on variable multi-graph and multimodal data for ASD diagnosis’, Liu et al. constructed a deep graph convolutional network (GCN) based on variable multi-graph and multimodal data (VMM-DGCN) for ASD diagnosis. Firstly, the functional connectivity matrix was constructed to extract primary features. Then, the authors constructed a variable multi-graph construction strategy to capture the multi-scale feature representations of each subject utilising convolutional filters with varying kernel sizes. Furthermore, the authors brought the non-imaging information into the feature representation at each scale. They constructed multiple population graphs based on multimodal data by fully considering the correlation between subjects. After extracting the deeper features of population graphs using the deep GCN (DeepGCN), the authors fused the node features of multiple subgraphs to perform node classification tasks for typical control and ASD patients. The proposed algorithm was evaluated on the autism brain imaging data exchange I (ABIDE I) dataset, achieving an accuracy of 91.62% and an area under the curve value of 95.74%. These results demonstrated its outstanding performance compared to other ASD diagnostic algorithms.

In the paper entitled ‘Safety control strategy of spinal lamina cutting based on force and cutting depth signals’, Zhang et al. innovatively believed the identification of the bone tissue status was regarded as a time series classification task. The classification algorithm LSTM-FCN was used to process fusion signals composed of force and cutting depth signals, thus accurately classifying the lamina bone tissue status. In addition, it was verified that the accuracy of the proposed method could reach 98.85% in identifying the state of porcine spinal laminectomy. The maximum penetration distance can be controlled within 0.6 mm, which is safe and can be used in practice.