Bill Lotter

@dfcidatascience and @harvardmed professor. Previously CTO & Co-Founder at DeepHealth (acq by RadNet). AI for medical imaging and neuroscience.

[Radiology: Artificial Intelligence @Radiology_AI]

Influence of Mammography Acquisition Parameters on #AI and Radiologist Interpretive Performance https://t.co/nZo7lY34BC @AMIAinformatics #AMIA2025 #AMIA25 #informatics #AI

[Radiology: Artificial Intelligence @Radiology_AI]

Distinguishing between Rigor and Transparency in FDA Marketing Authorization of AI-enabled Medical Devices https://t.co/cu6QYtnTV5 @AMIAinformatics #AMIA2025 #AMIA25 #informatics #AI

[Bill Lotter @bill_lotter]

Critically, image preprocessing and the handling of readily available parameters can be adjusted during AI development/deployment, offering a means for mitigating bias from an AI standpoint more generally. (8/n)

[Bill Lotter @bill_lotter]

Excited for "Acquisition parameters influence AI recognition of race in chest x-rays and mitigating these factors reduces underdiagnosis bias" to finally come out in @NatureComms https://t.co/N4iQJIq90x (1/n)

[Bill Lotter @bill_lotter]

There are many sources of bias and this approach is not sufficient, but it points to the importance of these "boring" factors and the potential of using AI to elucidate biases in clinical datasets to then reduce AI bias itself. (7/n)

[Wenxin (Vincent) Xu @VincentWenxinXu]

1/ Out now in @JCOCCI_ASCO! AI models can accurately impute performance status from clinical notes (AUROC 0.95), even when it's not explicitly documented. @kenlkehl @bill_lotter @DanaFarber_GU https://t.co/PpNuCCaEpE

[Cancer Discovery @CD_AACR]

New review in the May issue of Cancer Discovery—
Artificial intelligence (AI) in oncology: Current landscape, challenges, and future directions, by @bill_lotter, @ecerami et al.
https://t.co/HUaxFX62Ze
@dfcidatascience @harvardmed

[soria @jsoriamd]

The potential role of AI in clinical practice for oncology is anchored around improving detection, diagnosis, and treatment. https://t.co/xhODvLFAOW

[Bill Lotter @bill_lotter]

Our review on AI in Oncology is out today in @CD_AACR!

[Cancer Discovery @CD_AACR]

Just published for today's #AACR24 Plenary Session: #ArtificialIntelligence in Oncology—
Current Landscape, Challenges, and Future Directions, a review by @ecerami et al.
https://t.co/6c8oqQYjY7
@DanaFarber @dfcidatascience

[Dana-Farber Data Science @dfcidatascience]

We're pleased to announce @LorenzoTrippa's promotion to Professor, @DanaFarber & @HarvardBiostats . Dr. Trippa is an expert in developing and using statistical approaches to accelerate new discoveries in #cancer treatment.

https://t.co/D8Poc3cyJ4

[Stephanie McNamara @stephmcnamara]

Check out our work published this week in npj Digital Medicine! @bill_lotter @PaulYiMD

[npj Digital Medicine @npjDigitalMed]

A curated database of @US_FDA-cleared #ArtificialIntelligence products for medical imaging indicates a high prevalence of triage systems, variability in basic output characteristics, and often limited #explainability of AI predictions.

https://t.co/1h6ECKNuY0

[Sara Smolinski-Zhao, MD @SaraZhaoIRad]

Excited to announce our newly matched IR/DR residents! Welcome to the MGH IR family! @ShenNing_ @stephmcnamara @_PrateekGowda_ @mghradchiefs @SIRRFS #Match2024 #IRadres #futureradres

[Bill Lotter @bill_lotter]

New work with @stephmcnamara & @PaulYiMD. We study explainability & intended use in FDA-cleared AI for medical image interpretation. We create a curated database and find a high prevalence of triage devices, meaningful variation in basic outputs, and often limited explainability

[medRxiv @medrxivpreprint]

The clinician-AI interface: intended use and explainability in FDA-cleared AI devices for medical image interpretation https://t.co/MjOtmY4GBj #medRxiv

[Bill Lotter @bill_lotter]

Overall, our paper emphasizes the potential of using multimodal foundational models for intuitive, language-based explanations of visual tasks and dataset auditing (4)

[Bill Lotter @bill_lotter]

New paper with excellent (high school!) student @shoagarwal to be presented at ML4H. We use a pre-trained vision-language model to represent a new vision task as a linear combination of words: https://t.co/PzgSkF1Po7 (1)

[Bill Lotter @bill_lotter]

To functionally assess explainability, we perform a pilot reader study and find that the AI-identified words can help non-experts perform a medical task above chance levels (3)