Leveraging AI to facilitate clinical decision-making

Clusmann et al. describe how large language models such as ChatGPT could be used in medical practice, research and education. These models could democratize medical knowledge and facilitate access to healthcare, but there are also potential limitations to be considered.

Badal et al. outline principles that should be adopted during the development of artificial intelligence-based healthcare tools. These principles expand upon principles proposed by several organizations by emphasizing that AI be developed to improve longstanding health care challenges.

Koh, Papanikolaou et al. discuss the application of artificial intelligence in cancer imaging. The authors highlight opportunities for exploiting machine learning algorithms in this field, and outline barriers in their implementation and how these might be addressed.

Hu et al. describe their experiences running a training course for medical students about applying artificial intelligence to medical practice. They also provide recommendations for future training programs.

Vokinger et al. discuss potential sources of bias in machine learning systems used in medicine. The authors propose solutions to mitigate bias across the different stages of model development, from data collection and preparation to model evaluation and application.

Kundu discusses how artificial intelligence will transform medical practice and doctors’ training. The author explores the changing role of the clinician in the doctor-patient relationship, drawing parallels with the role of the pilot in light of increased automation in aviation.

Elsawy, Keenan, Chen et al. detect cataracts from color fundus photography using an explainable deep learning network called DeepOpacityNet. DeepOpacityNet detects cataracts more accurately than ophthalmologists and demonstrates that the absence of blood vessels is an indicator that cataracts are present.

Zhou et al. utilise deep learning to improve polygenic risk analysis for Alzheimer’s disease. Their computational approach outperforms existing statistical methods and helps to identify potential biological mechanisms of Alzheimer’s disease risk.

Steyaert, Qiu et al. develop a deep learning framework for multimodal data fusion for adult and pediatric brain tumors. Multimodal data models combining histopathology imaging and gene expression data outperform single data models in predicting prognosis.

Chang et al. classify people with Temporal Lobe Epilepsy (TLE), Alzheimer’s disease and healthy controls using a convolutional neural network algorithm applied to magnetic resonance imaging (MRI) scans. People with TLE can be distinguished, including those without easily identifiable TLE-associated MRI features.

D’Hondt et al. perform a qualitative and quantitative study on the implementation of machine learning (ML) in the intensive care unit (ICU). The authors interview hospital- and industry-based stakeholders to understand barriers in ML implementation and perform a number of ML experiments to quantify the impact of issues raised on model performance.

Ieki et al. train a deep learning model to estimate patients’ age from chest X-ray images. X-ray age is found to be an indicator of poor prognosis in patients with heart failure and patients admitted to the intensive care unit with cardiovascular disease.

Divard, Raynaud et al. compare artificial intelligence (AI)-based predictions of kidney allograft failure based on electronic health records with those made by transplant physicians of varying levels of experience. The ability of physicians to predict allograft failure is limited, with superior performance seen for the AI system.

Adam et al. evaluate the impact of biased AI recommendations on emergency decisions made by respondents to mental health crises. They find that descriptive rather than prescriptive recommendations made by the AI decision support system are more likely to lead to unbiased decision-making.

Gomes et al. develop machine learning models for gestational age and fetal malpresentation assessment on fetal ultrasound. The authors optimize their system for use in low-resource settings, using novice ultrasound operators, simplified imaging protocols, and low cost ultrasound devices.

Afrose, Song et al. highlight deficiencies in the widely accepted one-machine-learning-model-fits-all approach. The authors develop a bias correction method that produces specialized machine learning-based prognostication models for underrepresented racial and age groups.

Lutnick et al. develop a cloud-based deep learning tool for whole slide image segmentation. The authors provide several examples of its application in renal pathology, for segmenting glomeruli, interstitial fibrosis and other features of interest.

Inglese et al. develop a predictive model that computes multi-regional statistical morpho-functional mesoscopic traits from T1-weighted magnetic resonance imaging scans. Their model reliably discriminates people with Alzheimer’s disease-related pathologies from those without.

Pinckaers et al. develop a deep learning system to predict biochemical recurrence in prostate cancer patients treated with radical prostatectomy. The authors’ morphological biomarker provides predictive power beyond traditional Gleason grading, based on analysis of two clinical datasets from different institutions.

Liu et al. develop a deep learning-based tool to detect and segment diffusion abnormalities seen on magnetic resonance imaging (MRI) in acute ischemic stroke. The tool is tested in two clinical MRI datasets and outperforms existing algorithms in the detection of small lesions, potentially allowing clinicians and clinical researchers to more quickly and accurately diagnose and assess ischemic strokes.

Gamble and Jaroensri et al. develop deep learning systems to predict breast cancer biomarker status using H&E images. Their models enable slide-level and patch-level predictions for ER, PR and HER2, with interpretability analyses highlighting specific histological features associated with these markers.

Wulczyn et al. utilise a deep learning-based Gleason grading model to predict prostate cancer-specific mortality in a retrospective cohort of radical prostatectomy patients. Their model enables improved risk stratification compared to pathologists’ grading and demonstrates the potential for computational pathology in the management of prostate cancer.

Zhao et al. investigate optimal strategies to detect COVID-19 features in lung ultrasound images using deep neural networks trained with simulated and in vivo datasets. Including simulated data during training improves detection performance and training efficiency and is a promising alternative to curating thousands of patient images.

Nunez et al. investigate the use of natural language processing to predict which patients with cancer will see a psychiatrist or counselling using the initial oncology consultation document. Their study supports the use of such techniques with widely-available medical documents to better address the psychosocial needs of cancer patients.