Machine learning and artificial intelligence research for clinical medical image processing

National Library Of Medicine

Investigators

Linked publications, trials & patents

Abstract

1. Machine Learning and Artificial Intelligence for Biomedical Images Automated computer-aided diagnostic (CADx) tools driven by ML/AI methods based on deep learning (DL) are designed to detect and differentiate disease in medical images to improve automated disease prediction and add efficiencies to human performance. Toward this, we focused our research on various medical image analysis tasks such as quality assessment, image enhancement, region of interest detection and segmentation, image classification and prediction interpretation. Several advances were made to address these topics through applications to diseases of interest. Novel works done this year include image quality assessment for cervical and oral cavity images, echocardiography images and videos, and detection of Kaposi Sarcoma on dark-skinned population. Building on past years work, we not only made advances in deep learning networks and but also began to focus on AI prediction reliability. We recognize that training data characteristics significantly impact AI performance and special considerations must be made for generalizing the AI models so that they are robust across different imaging devices, population ethnicities, and disease severity expressed on the images. Further, understudies populations/ages can also impact the reliability of an AI. For these, we expanded our research on ensemble learning techniques which provide benefits from combining the predictions from different models and result in improved generalizability and overall accuracy. We also added generative AI for images to our work. This was explored to evaluate ability to detect COVID-19 on a chest x-ray dataset. We continued our interest measuring uncertainty in AI and correlating class confidence with diagnostic risk assessment 2. Disease-based ML/AI Research All software codes and data were made publicly available where possible. Tuberculosis: Automated segmentation of tuberculosis (TB)-consistent lesions in CXRs using DL methods can help reduce radiologist effort and supplement clinical decision-making. In the first study of its kind, we evaluated the benefits of using fine-grained annotations of TB-consistent lesions toward training ensembles for semantically segmenting TB-consistent lesions in both original and bone-suppressed frontal CXRs. Results showed that the stacking ensemble demonstrated superior segmentation performance. In a separate study, we investigated the benefits of selecting an appropriate loss function and quantifying uncertainty in predictions for segmenting TB manifestations in CXRs. Highly uncertain cases are referred to an expert thereby adding reliability to the classifier. We were the first to also analyze lateral CXRs using an ensemble of modality-specific convolutional neural networks (CNN) and vision transformer models (ViT) and obtained significantly superior performance which was verified using attention maps to highlight the discriminative image regions. Cardiovascular disease: Automated echocardiography (echo) analysis is benefited through use of machine learning for tasks such as image quality assessment, view classification, cardiac region segmentation, and quantification of diagnostic indices. Our novel and efficient DL-based real-time system for echo analysis and quantification is undergoing patent proection. Cervical cancer: Continuing research with NCI, we supported their efforts in inducing robustness in AI through training and evaluation based on Monte-Carlo methods. Oral cavity malignant lesion analysis: Oral cavity cancer is a common cancer that can result in significant impairments, and there is high mortality for the advanced stage. The final diagnosis is confirmed through histopathology, however high variability is observed among human experts in determining if a subject needs biopsy and identifying the correct biopsy location. Further, the disease can occur in different parts of the oral cavity. Toward developing an ML-based method that can help address these problems and reduce downstream classification errors, we automatically identify, with high accuracy, different anatomical sites in the oral cavity on the images that are verified using class activation maps obtained from both correct and incorrect predictions. This year we focused on two important problems: (i) Quality control - organizing and cleaning the dataset of over 4000 individuals such that it is free of duplicates, and missing or incorrectly labeled data; and, (ii) using computer-aided mechanisms for assisting in the annotation process. In this second effort, a small subset of the data in manually annotated and used for training a model. The model would be weakly trained as a result and likely to make errors. However, these errors can be corrected by a human judge. In a few iterations, the model is able to automatically annotate larger datasets and reduces the human workload to just verification. A separate effort ongoing is to detect mouth in the image to remove irrelevant background from consideration. Also of interest is detecting images with oral submucosal fibrosis (OSF) which are characterized by the clinician asking the patient to meaure how much their mouth opens. All of these narrow AI components would finally cascade into a framework for oral cancer AI. Kaposi Sarcoma on Dark-skinned population: This study focuses on the application of the state-of-art deep learning methodologies for Kaposi sarcoma (KS) lesion detection and classification on dark-skinned populations, aimed at two primary challenges: 1) The existing medical imaging databases lack a comprehensive representation of dark-skinned patients, consequently introducing bias into deep learning models; and 2) Limited research has been conducted on KS lesion detection and classification, particularly in dark-skin. The prevalent emphasis on fair-skinned populations not only compromises prediction accuracy but also erodes confidence in AI-generated predictions. This study seeks to enhance the reliability and inclusivity of AI applications within healthcare. As a first step, we developed methods to identify dark skin and separate it from irrelevant background in the image. We also evaluated several methods to automatically locate potential KS lesions and then classify as them as being KS or not. Further research in ongoing.