Background/Objectives: UNet-based models dominate medical image segmentation. Transformers have been added as an internal variant to these UNet-based architectures to improve feature learning. However, they have limitations in generalization and computational efficiency. Motivated by this idea, we have designed a two-stage novel hybrid segmentation framework, where tuners are added in cascade to the base architectures. Methods: Three sets of base UNets were designed, namely: B1: UNet1p, B2: UNet2p, and B3:UNet3p, and four sets of transformer-based tuners were designed, namely: T1:Transformer-augmented UNet, T2: Attention-guided UNet, T3: Swin Transformer-based UNet, and T4: Pyramid-based network, leading to 12 fused systems that combine three base UNets and four Tuners, namely: F1: B1 + T1, F2: B1 + T2, F3: B1 + T3, F4: B1 + T4; F5: B2 + T1, F6: B2 + T2, F7: B2 + T3, F8: B2 + T4, F9: B3 + T1, F10: B3 + T2, F11: B3 + T3, F12: B3 + T4. Results: The two-hybrid segmentation models are more effective and reliable than the single-stage UNet architecture. B3 + T4 achieved a Dice of 94.14% and Jaccard of 88.7%, surpassing prior baselines by 4.2% and 6.8%. It reduced cIMTE to 0.014 mm, a 36% improvement and the lowest reported to date, with cLIE and cMAE errors lowered by 40%. Conclusions: All 12 hybrid automated transformer-based models are highly accurate and reliable for wall segmentation in carotid ultrasound; they are a powerful paradigm for cardiovascular risk.
Single-cancer screening (SCS) methods improve survival in some cancers but miss ∼70% of cancer deaths from unscreened cancers. Multi-Cancer Detection (MCD) strategies aim to broaden early detection. We evaluate the MCD yield of screening WB-MRI (sWB-MRI) in a real-world clinical setting. In this retrospective single-center study, we reviewed sWB-MRI cases from 2022 with 12 months follow-up, excluding active cancer patients. The noncontrast, multiparametric sWB-MRI with Diffusion-Weighted-Imaging (1.5T) covered head to ankles. Radiologists provided structured reports for primary care physician (PCP) review. Follow-up data was gathered through direct-to-patient phone calls, focusing on capturing any interval history of biopsy and histopathologic outcomes. Of 1, 011 subjects (52% male, mean age 56 ± 28 years, median follow-up 14 ± 2 months), indications included proactive health (64%), general concerns (18%), and specific symptoms (18%). 9 radiologists reported the scans, with 5 reporting 90% (12%, 15%, 15%, 16%, & 32%). Results were reviewed with PCPs in 92.3% of cases. sWB-MRI led to targeted Tissue Sampling (TS) in 50 cases (4.9%). To isolate simple diagnostic-motivated TS (dTS) for further focused analysis, we excluded TS from clinically-indicated procedures beyond simple diagnostics: dTS = 41 (4.0%). Histopathology-confirmed Cancer Detection Percentages (CDP) with 95% CI were: Overall sWB-MRI CDP (O-CDP) = 2.2% (1.37-3.28%); dTS CDP = 51% (35-67%). O-CDP by age (# subjects, CDP %): <35 years (36, 0), 35-49 years (288, 1.4), 50-64 years (393, 2.6), 65-79 years (265, 3.0), >79 years (29, 0). Two false negatives (0.2%) were breast cancers. Of 22 detected cancers, 64% were retrospectively estimated localized, 36% regional/distant. 86% of cancers detected occurred in patients who did not indicate specific symptoms for their sWB-MRI reason. Of sWB-MRI detected cancer, 68% lacked SCS methods, while 32% had SCS options. Non-cancer clinically significant diagnoses (CSDs) included benign masses, aneurysms, liver disease, and pneumonia, prompting clinical action. Study limitations included reliance on patient-reported outcomes and lack of comprehensive medical records, limiting analysis of intermediate diagnostics and follow-up testing. sWB-MRI prompted pathologically-proven cancer diagnoses across diverse anatomical regions, including those outside standard SCS coverage. Prospective studies involving standardized sWB-MRI reporting frameworks (such as ONCO-RADS), larger cohorts, and designs that capture comprehensive intermediate diagnostic processes and long-term health outcomes, are essential to refine protocols and assess clinical validity (e.g., sensitivity, specificity, predictive values) and utility (e.g., patient outcomes, healthcare system impact). Candace Westgate, Rebecca Nayeri, Madhurima Datta, Jeffrey Venstrom, Rodrigo S. Pompa, Pratik Shingru, Saqib Basar, Sam Hashemi, Yosef Chodakiewitz. Noncontrast screening whole body MRI with diffusion-weighted imaging for multi cancer detection: a retrospective case series study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7406.
Individuals and their physicians receiving MCED tests may need to interpret the results they receive in the context of their particular cancer risks, both for positive (“cancer signal detected”) and for negative (“no cancer signal detected”) test results. Using estimated performance characteristics of an on-market MCED test we modeled post-test probabilities after receiving an MCED test result for a variety of typical populations stratified by five year age group and sex, and adjusted for smoking status (unadjusted, ever, former, never), as well as the duration of reduced cancer risk in those who tested negative. Using (1) 2006-2015 data (cancer incidence by age, sex) from the Surveillance, Epidemiology and End Reports (SEER) 17, (2) adjustments to incidence by smoking risk factors (3) test sensitivity from the third sub-study of the Circulating Cancer Genome Atlas, and (4) duration of detectable window from the American Cancer Society Cancer Prevention Study III and an exponential distribution for the detection window, we computed post-test probabilities of detection (for positive tests), and depletion of future cancers over various time intervals (for negative tests). Individuals receiving a negative MCED test result had 42-56% (min-max) the level of late stage (III+IV) cancer diagnosis in the year following blood draw, rising to 72-82% the second year if they did not return for an annual test. Individuals receiving a positive MCED test result with a cancer-signal origin (CSO) returned, had a cancer at that signal location if cancer is confirmed (97% of individual-CSO combinations), with sufficient positive predictive value at each CSO justifying workup (97% of individual-CSO combinations) with few exceptions, such as a breast CSO in males. An important complicating factor is the utilization of conventional screening or imaging tests which can significantly alter the pre-test probability of having certain cancers, for which we provide the example of colonoscopy and its effects over the ten-years post-negative screen. For common clinical covariates of age, sex, and smoking status this model provides important clinical context for interpreting test reports, including the potential for a significant increase in late stage, deadly interval cancers when testing at longer intervals than one year. Following the CSO estimate in the majority of individuals with positive test results is justified by comparison to usual clinical decision thresholds. Candace Westgate, Christina Clarke Dur, Eric Klein, Alpa V. Patel, Earl Hubbell. Estimated post-test probabilities of cancers for individuals receiving multi-cancer early detection (MCED) tests [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7132.
Background Cardiovascular diseases (CVD) cause 19 million fatalities each year and cost nations billions of dollars. Surrogate biomarkers are established methods for CVD risk stratification; however, manual inspection is costly, cumbersome, and error-prone. The contemporary artificial intelligence (AI) tools for segmentation and risk prediction, including older deep learning (DL) networks employ simple merge connections which may result in semantic loss of information and hence low in accuracy. Methodology We hypothesize that DL networks enhanced with attention mechanisms can do better segmentation than older DL models. The attention mechanism can concentrate on relevant features aiding the model in better understanding and interpreting images. This study proposes MultiNet 2.0 (AtheroPoint, Roseville, CA, USA), two attention networks have been used to segment the lumen from common carotid artery (CCA) ultrasound images and predict CVD risks. Results The database consisted of 407 ultrasound CCA images of both the left and right sides taken from 204 patients. Two experts were hired to delineate borders on the 407 images, generating two ground truths (GT1 and GT2). The results were far better than contemporary models. The lumen dimension (LD) error for GT1 and GT2 were 0.13±0.08 and 0.16±0.07 mm, respectively, the best in market. The AUC for low, moderate and high-risk patients’ detection from stenosis data for GT1 were 0.88, 0.98, and 1.00 respectively. Similarly, for GT2, the AUC values for low, moderate, and high-risk patient detection were 0.93, 0.97, and 1.00, respectively.The system can be fully adopted for clinical practice in AtheroEdge™ model by AtheroPoint, Roseville, CA, USA.
Biomedical image segmentation (BIS) task is challenging due to the variations in organ types, position, shape, size, scale, orientation, and image contrast. Conventional methods lack accurate and automated designs. Artificial intelligence (AI)-based UNet has recently dominated BIS. This is the first review of its kind that microscopically addressed UNet types by complexity, stratification of UNet by its components, addressing UNet in vascular vs. non-vascular framework, the key to segmentation challenge vs. UNet-based architecture, and finally interfacing the three facets of AI, the pruning, the explainable AI (XAI), and the AI-bias. PRISMA was used to select 267 UNet-based studies. Five classes were identified and labeled as conventional UNet, superior UNet, attention-channel UNet, hybrid UNet, and ensemble UNet. We discovered 81 variations of UNet by considering six kinds of components, namely encoder, decoder, skip connection, bridge network, loss function, and their combination. Vascular vs. non-vascular UNet architecture was compared. AP(ai)Bias 2.0-UNet was identified in these UNet classes based on (i) attributes of UNet architecture and its performance, (ii) explainable AI (XAI), and, (iii) pruning (compression). Five bias methods such as (i) ranking, (ii) radial, (iii) regional area, (iv) PROBAST, and (v) ROBINS-I were applied and compared using a Venn diagram. Vascular and non-vascular UNet systems dominated with sUNet classes with attention. Most of the studies suffered from a low interest in XAI and pruning strategies. None of the UNet models qualified to be bias-free. There is a need to move from paper-to-practice paradigms for clinical evaluation and settings.
