🚀 Congrats to all the accepted papers. Looking forward to seeing you all in Morocco. Submit final Camera Ready through OpenReview 🚀

Camera Read Submission FAQs: Please see the section here: ML-CDS @ MICCAI 2024 - FAQs (google.com)

🖊️ We are pleased to announce that the 12th ML-CDS workshop has been selected as an in-person event which will be co-located with MICCAI 2024!

🖊️ Join us in person at Marrakesh, Morocco on the 6th October, 2024 (8:30 - 13:00) .

Aims and Objectives: The goal of this workshop is to bring together machine learning and medical imaging researchers together with clinicians to discuss research that addresses how they are tackling the important challenges of acquiring and interpreting multimodality data for clinical decision support and treatment planning, and their adoption along with the latest developments in the field. In addition, due to the unique location of MICCAI this year in Africa, we would like to cover challenges as well as opportunities for clinical decision support in the developing world context, particularly, with respect to deployment in low cost rural settings, regulatory ease, as well as demographic tuning of models.

We are looking for original, high-quality submissions that address innovative research and development in the learning of multimodal medical data for use in clinical decision support and treatment planning. In addition, we are interested in soliciting submissions on techniques involving multi-modal image acquisition and reconstruction, novel methodologies and insights of multiscale multimodal medical images analysis, and empirical studies involving the application of multiscale multimodal imaging for clinical use.

Why Multimodal Learning for Clinical Decision Support? Diagnostic decision-making (using images and other modality data) is still very much an art for many physicians in their practices today due to a lack of quantitative tools and measurements. Deep learning for medical imaging showed initial promise for building clinical decision support systems. Furthermore, with medical images being acquired at multiple scales and/or multiple from modalities, multimodal fusion techniques have been increasingly applied in research studies and clinical practice to integrate and make sense of the patient data across scales of observation. With the advent of newer methods originating from large language models (LLMs) and generative multimodal models, newer possibilities arise for their adaptation and use in clinical decision support. However, their translation and adoption to clinical practice has still been slow with high expectations on accuracies for such systems in terms of both precision and recall as well as coverage. Regulatory approvals cover limited functionality, and restrict retraining of such systems on site which have also delayed adoption of them in hospitals.  

Multiple modalities of the data need to be analyzed to get a full picture of the patient’s conditions. These include images (x-ray, CT, MRI), videos and other time series, and textual data (free text reports and structured clinical data). In addition, with the routine availability of whole slide scanning technology, digital pathology data has become relevant. Additionally, for patient diagnosis and prognosis, some sort of “omics” (e.g. genomics, proteomics) data is also routinely obtained. All these provide the opportunity for multi-modal and multi-scale characterization of a patient’s disease profile. 

Analyzing these multimodal sources for disease-specific information across scales and across patients can reveal important similarities between patients and hence their underlying diseases and potential treatments. Researchers are now beginning to develop multimodal learning techniques on disease-specific information in modalities to find supporting evidence for a disease or to automatically learn associations of symptoms and their appearance in imaging. The role of clinical knowledge is also being actively explored. Large medical image collections are being offered for advancing research such as the recently released MIMIC and NIH image datasets for chest X-rays. However, accurate ground truth labeling of large scale datasets is proving to be a challenging problem.  While frameworks and tools for multiscale image analysis are still an open research question, facing the growing amount of data available from multiscale multimodal medical imaging facilities warrant new methods for the image analysis.