Towards Clinically Effective AI for Screening Mammography

Abstract Breast cancer is the most common cancer among women and a leading cause of cancer mortalityEarly detection of breast cancer can reduce mortality and morbiditywhich has led to widespread mammography screeningrecommended for women ageson a yearly or bi yearly basisReading the mammogram images to decide if cancer may be present is difficult due to the rarity of occurrencein a screening populationof women do not have cancerand the visual challenge of finding what can be a very subtle abnormality on a complex backgroundThis difficultycombined with the high volume of mammogramsmillion per year in the UShas led to a variety of proffered solutions including software known as computer aided diagnosisCADDespite early promisesuch solutions have not fulfilled their potential in improving outcomes and are largely thought to increase interpretation timesProductivity is increasingly a concern due to the rapidly growing use of digital breast tomosynthesisDBT orDmammographywhich has demonstrated higher cancer detection rates than traditionalD mammographybut also takes much longer to interpretAs a potential solutionthere has been significant interest in applying deep learning to mammographyDeep learningDLis a powerful field of machine learning which learns image features in an end to end fashion from dataand has been used to achieve human level performance on a number of imaging pattern recognition tasksThis proposal seeks to create DL based software for mammography that can be effective in a clinical setting throughaccurate and robust predictions on a diverse population of patientsinterpretable results from the DL modelnoblack boxanswersandapplicability to bothD mammography and DBTIn Phase Ithe aim is to improve model performance by training on additional data and incorporating additional algorithmic advancesPhase I will conclude with a clinical reader study comparing performance of the software to radiologistsIn Phase IIthe aim is to improve the clinical effectiveness of the software by automating quality detectionincorporating prior exams into the modeland expanding the training dataset to ensure results generalize to any woman eligible for screening mammographyThese improvements will apply to bothD and DBTAchieving the desired performance levels will enable a product that will improve productivity for radiologists and ensure consistent and accurate interpretations for patientsSuccess in this project would be a large step towards translating state ofthe art artificial intelligence to clinically effective tools for screening mammography Project Narrative Though mammography screening is widely considered beneficial for women agesthe varied challenges with reading the images lead to missed cancers or over diagnosiswhich in turn result in shorter length of life or unnecessary invasive procedures for many womenThe goal of this project is use recent breakthroughs in artificial intelligence to create a tool for doctors that will help them interpret mammograms more accurately and efficientlyThese improvements will enable better outcomes for women by improving access to quality mammography interpretationsespecially in resource constrained areas