arXiv subjects

Jake J. Kim

Publications and source records attributed to Jake J. Kim.

2 recordsLinked to original sources

arXiv2026

Characterizing the Radiation Dose to Measurement Accuracy Relationship across Multiple Metrics in Opportunistic Chest CT

Objectives: This study aims to characterize the dose-performance relationship for opportunistic CT and disentangle the contributions of segmentation failure and dose-dependent HU bias to performance degradation. Methods: Simulated low-dose CT images at 1-75% of full dose were generated from 50 paired full- and low-dose chest CT scans. An independent dataset of 22 paired PCCT acquisitions at lung cancer screening (LCS) and chest x-ray-equivalent (CXR) dose levels provided parallel real-world evaluation. Multiple quantitative disease metrics were obtained using deep learning-based segmentation followed by quantitative metric extraction. Classification performance was evaluated against full-dose reference standards, with additional analyses isolating the contributions of segmentation error and HU bias. Agreement between dose levels was assessed using Bland-Altman and correlation analyses. Results: Mean HU metrics maintained classification accuracy to CXR-equivalent dose (3%); bias correction improved accuracy from 88% to 96% for hepatic steatosis and from 84% to 90% for sarcopenia. Trabecular bone attenuation maintained 98% accuracy at LCS dose. Volume metrics (cardiomegaly) achieved 94% accuracy at CXR-equivalent dose. Threshold-based metrics required LCS dose for reliable classification; bias correction improved accuracy from 58% to 92%. Coronary artery calcification scoring reached 96% accuracy at LCS dose. In both Mayo and PCCT datasets, agreement analyses demonstrated strong correlation for all metrics except coronary artery calcification. Conclusions: Opportunistic CT is feasible at reduced dose levels though it becomes less robust at ultra-low doses. Distinct failure modes are caused by HU bias or segmentation failure and depend on the clinical task. Providers should be aware of these task-specific limitations when designing opportunistic screening programs.

physics.med-ph↗

arXiv2025

Deep learning approaches to surgical video segmentation and object detection: A Scoping Review

Introduction: Computer vision (CV) has had a transformative impact in biomedical fields such as radiology, dermatology, and pathology. Its real-world adoption in surgical applications, however, remains limited. We review the current state-of-the-art performance of deep learning (DL)-based CV models for segmentation and object detection of anatomical structures in videos obtained during surgical procedures. Methods: We conducted a scoping review of studies on semantic segmentation and object detection of anatomical structures published between 2014 and 2024 from 3 major databases - PubMed, Embase, and IEEE Xplore. The primary objective was to evaluate the state-of-the-art performance of semantic segmentation in surgical videos. Secondary objectives included examining DL models, progress toward clinical applications, and the specific challenges with segmentation of organs/tissues in surgical videos. Results: We identified 58 relevant published studies. These focused predominantly on procedures from general surgery [20(34.4%)], colorectal surgery [9(15.5%)], and neurosurgery [8(13.8%)]. Cholecystectomy [14(24.1%)] and low anterior rectal resection [5(8.6%)] were the most common procedures addressed. Semantic segmentation [47(81%)] was the primary CV task. U-Net [14(24.1%)] and DeepLab [13(22.4%)] were the most widely used models. Larger organs such as the liver (Dice score: 0.88) had higher accuracy compared to smaller structures such as nerves (Dice score: 0.49). Models demonstrated real-time inference potential ranging from 5-298 frames-per-second (fps). Conclusion: This review highlights the significant progress made in DL-based semantic segmentation for surgical videos with real-time applicability, particularly for larger organs. Addressing challenges with smaller structures, data availability, and generalizability remains crucial for future advancements.

eess.IV↗