Background: The use of artificial intelligence (AI) in the medical domain has attracted considerable research interest. Inference applications in the medical domain require energy-efficient AI models. In contrast to other types of data in visual AI, data from medical laboratories usually comprise features with strong signals. Numerous energy optimization techniques have been developed to relieve the burden on the hardware required to deploy a complex learning model. However, the energy efficiency levels of different AI models used for medical applications have not been studied.Objective: The aim of this study was to explore and compare the energy efficiency levels of commonly used machine learning algorithms—logistic regression (LR), k-nearest neighbor, support vector machine, random forest (RF), and extreme gradient boosting (XGB) algorithms, as well as four different variants of neural network (NN) algorithms—when applied to clinical laboratory datasets.Methods: We applied the aforementioned algorithms to two distinct clinical laboratory data sets: a mass spectrometry data set regarding Staphylococcus aureus for predicting methicillin resistance (3338 cases; 268 features) and a urinalysis data set for predicting Trichomonas vaginalis infection (839,164 cases; 9 features). We compared the performance of the nine inference algorithms in terms of accuracy, area under the receiver operating characteristic curve (AUROC), time consumption, and power consumption. The time and power consumption levels were determined using performance counter data from Intel Power Gadget 3.5.Results: The experimental results indicated that the RF and XGB algorithms achieved the two highest AUROC values for both data sets (84.7% and 83.9%, respectively, for the mass spectrometry data set; 91.1% and 91.4%, respectively, for the urinalysis data set). The XGB and LR algorithms exhibited the shortest inference time for both data sets (0.47 milliseconds for both in the mass spectrometry data set; 0.39 and 0.47 milliseconds, respectively, for the urinalysis data set). Compared with the RF algorithm, the XGB and LR algorithms exhibited a 45% and 53%-60% reduction in inference time for the mass spectrometry and urinalysis data sets, respectively. In terms of energy efficiency, the XGB algorithm exhibited the lowest power consumption for the mass spectrometry data set (9.42 Watts) and the LR algorithm exhibited the lowest power consumption for the urinalysis data set (9.98 Watts). Compared with a five-hidden-layer NN, the XGB and LR algorithms achieved 16%-24% and 9%-13% lower power consumption levels for the mass spectrometry and urinalysis data sets, respectively. In all experiments, the XGB algorithm exhibited the best performance in terms of accuracy, run time, and energy efficiency.Conclusions: The XGB algorithm achieved balanced performance levels in terms of AUROC, run time, and energy efficiency for the two clinical laboratory data sets. Considering the energy constraints in real-world scenarios, the XGB algorithm is ideal for medical AI applications.

Mycobacterium abscessus complex (MABC) has been reported to cause complicated infections. Subspecies identification of MABC is crucial for adequate treatment due to different antimicrobial resistance properties amid subspecies. However, long incubation days are needed for the traditional antibiotic susceptibility testing (AST). Delayed effective antibiotics administration often causes unfavorable outcomes. Thus, we proposed a novel approach to identify subspecies and potential antibiotic resistance, guiding early and accurate treatment. Subspecies of MABC isolates were determined by secA1, rpoB, and hsp65. Matrix-assisted laser desorption ionization-time of flight mass spectrometry (MALDI–TOF MS) spectra were analyzed, and informative peaks were detected by random forest (RF) importance. Machine learning (ML) algorithms were used to build models for classifying MABC subspecies based on spectrum. The models were validated by repeated five-fold cross-validation to avoid over-fitting. In total, 102 MABC isolates (52 subspecies abscessus and 50 subspecies massiliense) were analyzed. Top informative peaks including m/z 6715, 4739, etc. were identified. RF model attained AUROC of 0.9166 (95% CI: 0.9072–0.9196) and outperformed other algorithms in discriminating abscessus from massiliense. We developed a MALDI–TOF based ML model for rapid and accurate MABC subspecies identification. Due to the significant correlation between subspecies and corresponding antibiotics resistance, this diagnostic tool guides a more precise and timelier MABC subspecies-specific treatment.

Cervical cancer, a major health concern among women worldwide, is closely linked to human papillomavirus (HPV) infection. This study explores the evolving landscape of HPV molecular epidemiology in Taiwan over a decade (2010–2020), where prophylactic HPV vaccination has been implemented since 2007. Analyzing data from 40,561 vaginal swab samples, with 42.0% testing positive for HPV, we reveal shifting trends in HPV genotype distribution and infection patterns. The 12 high-risk genotypes, in order of decreasing percentage, were HPV 52, 58, 16, 18, 51, 56, 39, 59, 33, 31, 45, and 35. The predominant genotypes were HPV 52, 58, and 16, accounting for over 70% of cases annually. The proportions of high-risk and non-high-risk HPV infections varied across age groups. High-risk infections predominated in sexually active individuals aged 30-50 and were mixed-type infections. The composition of high-risk HPV genotypes was generally stable over time; however, HPV31, 33, 39, and 51 significantly decreased over the decade. Of the strains, HPV31 and 33 are shielded by the nonavalent HPV vaccine. However, no reduction was noted for the other seven genotypes. This study offers valuable insights into the post-vaccine HPV epidemiology. Future investigations should delve into HPV vaccines’ effects and their implications for cervical cancer prevention strategies. These findings underscore the need for continued surveillance and research to guide effective public health interventions targeting HPV-associated diseases.

Capillary zone electrophoresis–immunosubtraction (CZE-IS) is an essential laboratory test in diagnosing plasma cell neoplasms. However, the current interpretation of the test results is subjective. To evaluate CZE-IS in a more precise manner, this study proposed five key indexes, namely sharpness index, light chain index, immunoglobulin G index, immunoglobulin A index, and immunoglobulin M index. The reference intervals of these indexes were established using CZE-IS curve data from a clinical laboratory of a referral medical center. A total of 1000 cases with normal electrophoretic patterns were sampled for reference intervals establishment, and an additional 20 cases were included for validation. The following reference intervals in the γ zone were established: 1-6 (sharpness index), 1.06-2.71 (light chain index), 37-454 (immunoglobulin G index), (−9)-41 (immunoglobulin A index), and (−16)-46 (immunoglobulin M index). For the β2 zone, the reference intervals were 3-17 (sharpness index), 0.44-1.90 (light chain index), (−7)-61 (immunoglobulin G index), 2-117 (immunoglobulin A index), and (−12)-35 (immunoglobulin M index). The diagnostic performance of reference intervals of the proposed indexes in validation ranged from 95% to 100%. CZE-IS indexes provide the objective quantification of key characteristics of CZE-IS curves and improve the precision of CZE-IS interpretation.

Background: Capillary blood sampling may improve access to multi-cancer early detection (MCED) by enabling less invasive and more decentralized specimen collection. However, its paired-specimen measurement agreement with conventional venous blood for protein biomarker-based MCED testing requires validation. Objective: The objective of this study was to evaluate agreement between capillary and venous blood specimens for OneTest protein biomarkers and to assess the exact risk category agreement of the OneTest Premium algorithm outputs generated from both specimen types. Methods: This two-stage study compared paired venous and capillary blood specimens. Study 1 enrolled 143 Taiwanese adults undergoing clinically indicated biomarker testing to evaluate paired-specimen measurement agreement. Study 2 involved an exploratory paired-specimen evaluation in 13 asymptomatic U.S. volunteers using an upper arm capillary blood collection device. Biomarker concentrations were compared using paired statistical testing, correlation analysis, and Bland–Altman analysis. OneTest Premium risk categories and cancer type prediction outputs were compared between specimen types. Results: In Study 1, no statistically significant differences were observed between capillary and venous specimens for any evaluated biomarker. Descriptive scatterplots showed strong linear associations between paired capillary and venous biomarker measurements across the observed concentration ranges. Bland–Altman analysis showed mean biases close to zero for most biomarkers, with biomarker-specific limits of agreement. In Study 2, all 12 biomarkers were successfully measured in all 13 participants. OneTest Premium risk categorization was concordant in 11 of 13 participants; discordant cases involved adjacent low-risk and mild-risk categories, with no major or very major classification errors. Given the small sample size, these findings are descriptive and preliminary. Conclusions: Capillary blood specimens showed paired measurement agreement with venous specimens in Study 1. The exploratory findings from Study 2 suggest that capillary-derived measurements may support the generation of OneTest Premium outputs.

Background: Machine learning models are increasingly applied in clinical practice. However, conventional performance metrics such as sensitivity, specificity, and accuracy do not capture the structural characteristics of predictive probability distributions, limiting their utility for optimizing decision thresholds and supporting clinical decision-making. Methods: We propose the D-index, a novel metric designed to quantify asymmetry in predicted probability distributions by measuring the difference in skewness between positive and negative predictions. The D-index was evaluated using simulated datasets with skewness-dominant and kurtosis-dominant distributions at three performance levels (AUROC = 0.7, 0.8, and 0.9). Its practical utility was further validated using real-world MALDI-TOF mass spectrometry data for predicting vancomycin resistance in Enterococcus faecium, comparing three machine learning algorithms: deep learning, XGBoost, and random forest. Results: In simulated data, the D-index increased with model performance only in skewness-dominant distributions, demonstrating selective sensitivity to distributional asymmetry. In real-world data, the D-index distinguished models with similar conventional performance but different distributional structures. Models with higher D-index values retained a larger proportion of confidently classified cases while achieving meaningful accuracy improvements under a gray-zone strategy. In contrast, models with lower D-index values required exclusion of more cases to achieve comparable or greater accuracy gains. Conclusions: The D-index provides complementary information beyond traditional performance metrics by capturing structural properties of predictive distributions. It enables identification of models that are better suited for gray-zone strategies and supports more reliable clinical decision-making. This distribution-aware approach offers a practical tool for improving the safety, efficiency, and clinical utility of machine learning models in biomedical applications.