In the rapidly evolving field of medical image analysis utilizing artificial intelligence (AI), the selection of appropriate computational models is critical for accurate diagnosis and patient care. This literature review provides a comprehensive comparison of vision transformers (ViTs) and convolutional neural networks (CNNs), the two leading techniques in the field of deep learning in medical imaging. We conducted a survey systematically. Particular attention was given to the robustness, computational efficiency, scalability, and accuracy of these models in handling complex medical datasets. The review incorporates findings from 36 studies and indicates a collective trend that transformer-based models, particularly ViTs, exhibit significant potential in diverse medical imaging tasks, showcasing superior performance when contrasted with conventional CNN models. Additionally, it is evident that pre-training is important for transformer applications. We expect this work to help researchers and practitioners select the most appropriate model for specific medical image analysis tasks, accounting for the current state of the art and future trends in the field.

In recent years, advances in artificial intelligence (AI) technology have led to the rapid clinical implementation of devices with AI technology in the medical field. More than 60 AI-equipped medical devices have already been approved by the Food and Drug Administration (FDA) in the United States, and the active introduction of AI technology is considered to be an inevitable trend in the future of medicine. In the field of oncology, clinical applications of medical devices using AI technology are already underway, mainly in radiology, and AI technology is expected to be positioned as an important core technology. In particular, “precision medicine,” a medical treatment that selects the most appropriate treatment for each patient based on a vast amount of medical data such as genome information, has become a worldwide trend; AI technology is expected to be utilized in the process of extracting truly useful information from a large amount of medical data and applying it to diagnosis and treatment. In this review, we would like to introduce the history of AI technology and the current state of medical AI, especially in the oncology field, as well as discuss the possibilities and challenges of AI technology in the medical field.

Lung cancer is one of the leading causes of death worldwide. Therefore, understanding the factors linked to patient survival is essential. Recently, multi-omics analysis has emerged, allowing for patient groups to be classified according to prognosis and at a more individual level, to support the use of precision medicine. Here, we combined RNA expression and miRNA expression with clinical information, to conduct a multi-omics analysis, using publicly available datasets (the cancer genome atlas (TCGA) focusing on lung adenocarcinoma (LUAD)). We were able to successfully subclass patients according to survival. The classifiers we developed, using inferred labels obtained from patient subtypes showed that a support vector machine (SVM), gave the best classification results, with an accuracy of 0.82 with the test dataset. Using these subtypes, we ranked genes based on RNA expression levels. The top 25 genes were investigated, to elucidate the mechanisms that underlie patient prognosis. Bioinformatics analyses showed that the expression levels of six out of 25 genes (ERO1B, DPY19L1, NCAM1, RET, MARCH1, and SLC7A8) were associated with LUAD patient survival (p < 0.05), and pathway analyses indicated that major cancer signaling was altered in the subtypes.

Among diffuse gliomas, oligodendrogliomas show relatively better prognosis, respond well to radiotherapy and chemotherapy, and seldom progress to very aggressive tumors. To elucidate the genetic and epigenetic background for such behavior and tumor evolution during tumor relapse, we comparatively analyzed 12 pairs of primary and recurrent oligodendrogliomas with 1p/19q-codeletion. Initial treatment for these patients was mostly chemotherapy alone. Temozolomide was used for 3, and procarbazine, nimustine and vincristine (PAV chemotherapy) were used for 7 patients. World Health Organization histological grade at recurrence was mostly stable; it was increased in 2, the same in 9, and decreased in 1 cases. Whole-exome sequencing demonstrated that the rate of shared mutation between the primary and recurrent tumors was relatively low, ranging from 3.2-57.9% (average, 33.3%), indicating a branched evolutionary pattern. The trunk alterations that existed throughout the course were restricted to IDH1 mutation, 1p/19q-codeletion, and TERT promoter mutation, and mutation of the known candidate tumor suppressor genes CIC and FUBP1 were not consistently observed between primary and recurrent tumors. Multiple sampling from different regions within a tumor showed marked intratumoral heterogeneity. Notably, in general, the number of mutations was not significantly different after recurrence, remaining under 100, and no hypermutator phenotype was observed. FUBP1 mutation, loss of chr. 9p21, and TCF12 mutation were among a few recurrent de novo alterations that were found at recurrence, indicating that these events were clonally selected at recurrence but were not enough to enhance malignancy. Genome-wide methylation status, measured by Illumina 450 K arrays, was stable between recurrence and the primary tumor. In summary, although oligodendroglioma displays marked mutational heterogeneity, histological malignant transformation accompanying events such as considerable increase in mutation number and epigenetic profile change were not observed at recurrence, indicating that noticeable temporal and spatial genetic heterogeneity in oligodendrogliomas does not result in rapid tumor progression.

Upper gastrointestinal cancer (UGC) sometimes metastasizes to the splenic hilum lymph node (SHLN). However, surgical removal of SHLN is technically difficult, and the risk of postoperative complications is high. Although there are models that predict SHLN metastasis, they usually only provide point estimates of risk, and there is a lack of sufficient information. To address this issue, we aimed to develop a Bayesian logistic regression model called Bayes-SHLNM. The performance of the models was compared with that of the frequentist logistic regression (FLR) model as a benchmark, and the posterior probability distribution (PPD) was shown individually. The performance of Bayes-SHLNM was equivalent to that of the FLR model, and the PPD for each case was visualized as the uncertainty. These results indicate that the Bayes-SHLNM model has the potential to be used as a decision support system in clinical settings where uncertainty is high.

Fetal cardiac ultrasound screening is vital for the prenatal diagnosis of congenital heart disease (CHD). However, understanding complex cardiovascular structures and continuity from two-dimensional (2D) ultrasound images remains challenging. Therefore, this study developed and evaluated an artificial intelligence (AI)-based three-dimensional (3D) visualization approach that reconstructs fetal cardiovascular structures from 2D ultrasound sweep video. To this end, information from the adjacent frames was used. Instead of interpreting each ultrasound frame independently, this method incorporates neighboring frames, enabling the continuous visualization of the right- and left-sided functional circulatory pathways. Our method achieved a mean Dice coefficient of 0.785 on an independent test dataset, demonstrating that the adjacent frame information can provide stable 3D segmentation. Diffusion-based anatomical plausibility analysis trained exclusively on normal anatomies showed exploratory discrimination between normal fetuses and fetuses with CHD (AUC, 0.861; 95% CI, 0.695–0.980). In a clinical comparison study involving 12 obstetricians, the introduction of this AI-driven 3D cardiovascular visualization slightly improved the screening accuracy from 0.779 to 0.808 and significantly improved the confidence-weighted accuracy score (p = 0.002). Our approach can potentially support examiners and facilitate fetal cardiac ultrasound screenings.