Last updated 11:30 a.m.
Today, more than 1,700 senior health care leaders have gathered for a global conference featuring speakers from the Brigham and elsewhere. This year’s World Medical Innovation Forum will focus on Artificial Intelligence (AI) and its growing impact on clinical care. Brigham Clinical & Research News is here, bringing you live updates from “First Look,” a series of rapid fire presentations by early career investigators about the applications of AI in clinical care.
Brigham Presenters Kick Off 2019 World Medical Innovation Forum with Fresh Ideas for AI at First Look Session
Brigham early-career investigators are taking the stage this morning to share their bold ideas and big projects at the intersection of artificial intelligence and health care. Giles Boland, MD, chair of the Brigham’s Department of Radiology and Trung Do of Partners HealthCare, serve as this morning’s moderators.
Sandro Santagata, MD, PhD: “Multiplexed Tissue Imaging and Quantitative Pathology for Discovery and Translational Medicine”
Santagata is a clinical sub-specialist in Neuropathology with a background in Neuroscience and Pathology. His research focuses on identifying therapeutic targets in brain tumors and implementing new technologies for tumor imaging. He is currently leading efforts to develop multi-dimensional tumor atlases as part of the Biden Moonshot Human Tumor Atlas Network (HTAN).
- “Our hope is that this will be used to reveal important vulnerabilities in tumors that will ultimately benefit our patients.”
- For decades, diagnosing cancers involved revising histology sections. New genomic technologies have expanded our diagnostic capabilities — genetic mutations in cancer oncogenes can now be identified and subsequently targeted using cancer immunotherapies.
- Despite the advances in targeted immunotherapies, many patients do not have tumors which are responsive to this intervention. New analytical approaches are needed to study tumors.
- Santagata’s work focuses on developing novel computation and analytic methods using mass spectrometry (MS). The team is conducting the first clinical trial to classify new data and samples using MS to identify tumor grade, tumor prognosis, and drugs present in tumor cells
- These methods represent tremendous opportunity — including disrupting the current paradigm for intraoperative diagnosis and allowing for surgeons to cut down on OR time.
- But there is one downside: the resolution with MS is not as good as with traditional methods, such as MRI. The team has taken on a new approach for high resolution pathology called tissue cyclic immunofluorescence. In time, with AI and other approaches, features such as nuclear size and cell shape can be extracted to define new tumor cell states
Tina Kapur, PhD: “Using AI to Better Visualize Needles in Ultrasound-Guided Liver Biopsies”
Kapur is the executive director of Image-Guided Therapy in the Department of Radiology. Kapur’s research focuses on medical image computing and computer-aided interventions in domains of neurosurgery, surgical navigation, and MR-guided pelvic brachytherapy. Kapur also leads initiatives in open science, and is the founding director of the first open-science hackathons for medical image computing, “NA-MIC Project Week.”
- The big picture problem Kapur is interested in is using AI to auto detect any man-made object from any medical image: stents, implants, surgical instruments and needles.
- Nearly 1 million ultrasound guided liver biopsies/year, increasing at a rate of 4 percent
- How can we use deep learning tech from AI to find tips of needles from ultrasound images to make liver biopsies safer for patients and easier for the physician?
- The team is finding needles in image hay stacks. They want to enhance visualization of the needle using the power of AI.
- We want to have a needle button that shows a visualized needle whenever the physician wants to see it, says Kapur. We want to beat the competition. Ours is not a navigation system or guides/coils. Ours comes with no added time, complexity, incremental cost.
Li Zhou, MD, PhD: “Machine Learning and NLP to Track Disease Progression and Predict Health Outcomes”
Zhou is a lead investigator at the Division of General Internal Medicine and Primary Care. Zhou’s research focuses on the application of various subfields of AI including natural language processing and machine learning to various clinical domains, such as phenotyping, medication reconciliation, and adverse drug reaction detection. Currently, Zhou is developing innovative methods for error detection in clinical notes generated by speech software, and using machine learning to extract knowledge from malpractice claims to enhance coding and analytics for risk management.
- “Society is changing, and more patients are now living with chronic illnesses. There is an urgent need for increased depth and breadth in palliative care delivery, and timely referral to pall care can improve patient well-being.”
- One of the biggest challenges in palliative care is identifying patients who can benefit from care and the right time to initiate interventions. There is an urgent need to leverage information technology and electronic health record (EHR) data towards guiding palliative interventions.
- In a 2018 study, Zhou et al. studied dementia patients over the last years of life. Natural language processing was used to capture themes mentioned in EHR notes over this time.
- They then developed a deep learning neural network system to predict mortality with longitudinal clinical notes. Preliminary work shows promising results in predicting two-year mortality for dementia patients – performing better than clinician predictions.
- Next, Zhou’s team hopes to enhance the model by inputting other data sources and to put the tool into clinical practice.
Vesela Kovacheva, MD, PhD: “Harnessing the Power of Machine Learning to Automate Drug Infusions in the OR and ICU”
Kovacheva is an attending clinician in the department of Anesthesiology, with a focus on obstetric and reproductive medicine anesthesiology. Kovacheva’s research focuses on creating algorithms of drug infusion using deep learning methods, hoping to integrate data into clinical care.
- As an expert in obstetric anesthesiology, Kovacheva helps a lot of mothers deliver their babies safely and comfortably at the Brigham. She is responsible for, among other things, titrating drug infusions for various drugs, including drugs known as vasopresors that keep blood pressure normal
- The decision to administer vasopressors is taken every minute based on the blood pressure.
- Kacheva is working with partners to use real time data fed into a machine learning algorithm to predict how much medication the patient will need in the next moment. This will augment and be supervised by the physician
- They plan to pilot with blood pressure medications during c section and have laid groundwork for a phase 1 clinical trial.
- “For me, this would be an amazing super power.”
Chris Sidey-Gibbons, PhD: “Three Computational Techniques and One Tool to Bring the Patient Voice into Care”
Sidey-Gibbons is the co-founder and co-director of the Brigham’s Patient-Reported Outcomes, Value, and Experience (PROVE) center. Using computational tools, Sidey-Gibbons aims to assess patient outcomes including quality of life, fatigue, and depression to ultimately transform the collection, analysis, and reporting of such variables in surgery. His recent work focuses on using natural language processing algorithms to parse unstructured patient reports.
- “I’m passionate about patient reported data.”
- Patient-reported data is usually collected and used to improve communication, identification and treatment, quality of life, mental health, satisfaction, and survival. However, this information is currently collected on a piece of paper on an electronic screen – performing the same operations as it has in the 1960s.
- To improve the nuance and quality of patient-reporting interventions, Sidey-Gibbons’ lab has aimed to reduce response burden through computer adaptive assessment, a platform which mimics an intelligent doctor and allows for shortening of the questionnaire, and increased accuracy.
- Supported by funding from the UK and the NIH, the team designed a program to draw insights from open, unstructured data.
- Team has been developing software, Concerto, which is an open-source software which allows clinicians to use CAT, machine learning and feedback to improve communication. They are currently working to develop two additional systems, imPROVE and INSPiRES.
Bharti Khurana, MD: “Making the Invisible Visible: Bringing Intimate Partner Violence into Focus”
Khurana, clinical director for Emergency Musculoskeletal Radiology, is an expert in the field of orthopedic trauma imaging. Her research focuses on using radiology and machine learning to detect victims of intimate partner violence and harnessing the powers of AI to move beyond clinical diagnosis.
- The most dangerous place around the world for women is home. 50k homicides committed by a woman’s partner or family members.
- 55 percent of female homicides are linked to intimate partner violence
- If we can diagnose early, we can intervene and prevent IPV
- Radiologists can create an objective, unbiased report. Khurana and colleagues published a pilot study earlier this year on using radiology data to find imaging patterns specific to IPV.
- Team is developing automated identification of IPV to provide clinical decision support to predict risk probability.
Are you attending this year’s forum? Leave us a comment below! Want to learn more about WMIF 2019? Visit here.