Recent advances on loss functions in deep learning for computer vision

Lightweight cotton diseases real-time detection model for resource-constrained devices in natural environments

Cotton, a vital textile raw material, is intricately linked to people's livelihoods. Throughout the cotton cultivation process, various diseases threaten cotton crops, significantly impacting both cotton quality and yield. Deep learning has emerged as a crucial tool for detecting these diseases. However, deep learning models with high accuracy often come with redundant parameters, making them challenging to deploy on resource-constrained devices. Existing detection models struggle to strike the right balance between accuracy and speed, limiting their utility in this context. This study introduces the CDDLite-YOLO model, an innovation based on the YOLOv8 model, designed for detecting cotton diseases in natural field conditions. The C2f-Faster module replaces the Bottleneck structure in the C2f module within the backbone network, using partial convolution. The neck network adopts Slim-neck structure by replacing the C2f module with the GSConv and VoVGSCSP modules, based on GSConv. In the head, we introduce the MPDIoU loss function, addressing limitations in existing loss functions. Additionally, we designed the PCDetect detection head, integrating the PCD module and replacing some CBS modules with PCDetect. Our experimental results demonstrate the effectiveness of the CDDLite-YOLO model, achieving a remarkable mean average precision (mAP) of 90.6%. With a mere 1.8M parameters, 3.6G FLOPS, and a rapid detection speed of 222.22 FPS, it outperforms other models, showcasing its superiority. It successfully strikes a harmonious balance between detection speed, accuracy, and model size, positioning it as a promising candidate for deployment on an embedded GPU chip without sacrificing performance. Our model serves as a pivotal technical advancement, facilitating timely cotton disease detection and providing valuable insights for the design of detection models for agricultural inspection robots and other resource-constrained agricultural devices.

A machine learning and deep learning-based integrated multi-omics technique for leukemia prediction

In recent years, scientific data on cancer has expanded, providing potential for a better understanding of malignancies and improved tailored care. Advances in Artificial Intelligence (AI) processing power and algorithmic development position Machine Learning (ML) and Deep Learning (DL) as crucial players in predicting Leukemia, a blood cancer, using integrated multi-omics technology. However, realizing these goals demands novel approaches to harness this data deluge. This study introduces a novel Leukemia diagnosis approach, analyzing multi-omics data for accuracy using ML and DL algorithms. ML techniques, including Random Forest (RF), Naive Bayes (NB), Decision Tree (DT), Logistic Regression (LR), Gradient Boosting (GB), and DL methods such as Recurrent Neural Networks (RNN) and Feedforward Neural Networks (FNN) are compared. GB achieved 97 % accuracy in ML, while RNN outperformed by achieving 98 % accuracy in DL. This approach filters unclassified data effectively, demonstrating the significance of DL for leukemia prediction. The testing validation was based on 17 different features such as patient age, sex, mutation type, treatment methods, chromosomes, and others. Our study compares ML and DL techniques and chooses the best technique that gives optimum results. The study emphasizes the implications of high-throughput technology in healthcare, offering improved patient care.

Toward interpretable and generalized mitosis detection in digital pathology using deep learning

Objective

The mitotic activity index is an important prognostic factor in the diagnosis of cancer. The task of mitosis detection is difficult as the nuclei are microscopic in size and partially labeled, and there are many more non-mitotic nuclei compared to mitotic ones. In this paper, we highlight the challenges of current mitosis detection pipelines and propose a method to tackle these challenges.

Methods

Our proposed methodology is inspired from recent research on deep learning and an extensive analysis on the dataset and training pipeline. We first used the MiDoG'22 dataset for training, validation, and testing. We then tested the methodology without fine-tuning on the TUPAC'16 dataset and on a real-time case from Shaukat Khanum Memorial Cancer Hospital and Research Centre.

Results

Our methodology has shown promising results both quantitatively and qualitatively. Quantitatively, our methodology achieved an F1-score of 0.87 on the MiDoG'22 dataset and an F1-score of 0.83 on the TUPAC dataset. Qualitatively, our methodology is generalizable and interpretable across various datasets and clinical settings.

Conclusion

In this paper, we highlight the challenges of current mitosis detection pipelines and propose a method that can accurately predict mitotic nuclei. We illustrate the accuracy, generalizability, and interpretability of our approach across various datasets and clinical settings. Our methodology can speed up the adoption of computer-aided digital pathology in clinical settings.

Enhancing percutaneous coronary intervention with heuristic path planning and deep-learning-based vascular segmentation

Percutaneous coronary intervention (PCI) is a minimally invasive technique for treating vascular diseases. PCI requires precise and real-time visualization and guidance to ensure surgical safety and efficiency. Existing mainstream guiding methods rely on hemodynamic parameters. However, these methods are less intuitive than images and pose some challenges to the decision-making of cardiologists. This paper proposes a novel PCI guiding assistance system by combining a novel vascular segmentation network and a heuristic intervention path planning algorithm, providing cardiologists with clear and visualized information. A dataset of 1077 DSA images from 288 patients is also collected in clinical practice. A Likert Scale is also designed to evaluate system performance in user experiments. Results of user experiments demonstrate that the system can generate satisfactory and reasonable paths for PCI. Our proposed method outperformed the state-of-the-art baselines based on three metrics (Jaccard: 0.4091, F1: 0.5626, Accuracy: 0.9583). The proposed system can effectively assist cardiologists in PCI by providing a clear segmentation of vascular structures and optimal real-time intervention paths, thus demonstrating great potential for robotic PCI autonomy.

Federated Learning in Computational Toxicology: An Industrial Perspective on the Effiris Hackathon

In silico approaches have acquired a towering role in pharmaceutical research and development, allowing laboratories all around the world to design, create, and optimize novel molecular entities with unprecedented efficiency. From a toxicological perspective, computational methods have guided the choices of medicinal chemists toward compounds displaying improved safety profiles. Even if the recent advances in the field are significant, many challenges remain active in the on-target and off-target prediction fields. Machine learning methods have shown their ability to identify molecules with safety concerns. However, they strongly depend on the abundance and diversity of data used for their training. Sharing such information among pharmaceutical companies remains extremely limited due to confidentiality reasons, but in this scenario, a recent concept named "federated learning" can help overcome such concerns. Within this framework, it is possible for companies to contribute to the training of common machine learning algorithms, using, but not sharing, their proprietary data. Very recently, Lhasa Limited organized a hackathon involving several industrial partners in order to assess the performance of their federated learning platform, called "Effiris". In this paper, we share our experience as Roche in participating in such an event, evaluating the performance of the federated algorithms and comparing them with those coming from our in-house-only machine learning models. Our aim is to highlight the advantages of federated learning and its intrinsic limitations and also suggest some points for potential improvements in the method.

Deep Learning Techniques with Genomic Data in Cancer Prognosis: A Comprehensive Review of the 2021-2023 Literature

Deep learning has brought about a significant transformation in machine learning, leading to an array of novel methodologies and consequently broadening its influence. The application of deep learning in various sectors, especially biomedical data analysis, has initiated a period filled with noteworthy scientific developments. This trend has majorly influenced cancer prognosis, where the interpretation of genomic data for survival analysis has become a central research focus. The capacity of deep learning to decode intricate patterns embedded within high-dimensional genomic data has provoked a paradigm shift in our understanding of cancer survival. Given the swift progression in this field, there is an urgent need for a comprehensive review that focuses on the most influential studies from 2021 to 2023. This review, through its careful selection and thorough exploration of dominant trends and methodologies, strives to fulfill this need. The paper aims to enhance our existing understanding of applications of deep learning in cancer survival analysis, while also highlighting promising directions for future research. This paper undertakes aims to enrich our existing grasp of the application of deep learning in cancer survival analysis, while concurrently shedding light on promising directions for future research in this vibrant and rapidly proliferating field.

Health-Aware Food Recommendation Based on Knowledge Graph and Multi-Task Learning

Current food recommender systems tend to prioritize either the user's dietary preferences or the healthiness of the food, without considering the importance of personalized health requirements. To address this issue, we propose a novel approach to healthy food recommendations that takes into account the user's personalized health requirements, in addition to their dietary preferences. Our work comprises three perspectives. Firstly, we propose a collaborative recipe knowledge graph (CRKG) with millions of triplets, containing user-recipe interactions, recipe-ingredient associations, and other food-related information. Secondly, we define a score-based method for evaluating the healthiness match between recipes and user preferences. Based on these two prior perspectives, we develop a novel health-aware food recommendation model (FKGM) using knowledge graph embedding and multi-task learning. FKGM employs a knowledge-aware attention graph convolutional neural network to capture the semantic associations between users and recipes on the collaborative knowledge graph and learns the user's requirements in both preference and health by fusing the losses of these two learning tasks. We conducted experiments to demonstrate that FKGM outperformed four competing baseline models in integrating users' dietary preferences and personalized health requirements in food recommendations and performed best on the health task.

Novel Methodology to Recover Road Surface Height Maps from Illuminated Scene through Convolutional Neural Networks

Road surface properties have a major impact on pavement's life service conditions. Nowadays, contactless techniques are widely used to monitor road surfaces due to their portability and high precision. Among the different possibilities, laser profilometers are widely used, even though they have two major drawbacks: spatial information is missed and the cost of the equipment is considerable. The scope of this work is to show the methodology used to develop a fast and low-cost system using images taken with a commercial camera to recover the height information of the road surface using Convolutional Neural Networks. Hence, the dataset was created ad hoc. Based on photometric theory, a closed black-box with four light sources positioned around the surface sample was built. The surface was provided with markers in order to link the ground truth measurements carried out with a laser profilometer and their corresponding intensity values. The proposed network was trained, validated and tested on the created dataset. Three loss functions where studied. The results showed the Binary Cross Entropy loss to be the most performing and the best overall on the reconstruction task. The methodology described in this study shows the feasibility of a low-cost system using commercial cameras based on Artificial Intelligence.