In daily activities, the foot, as a biomechanical structure, bears the body weight and plays a crucial role in human locomotion. When standing or walking, the ground reaction force applied to the foot can reach 1.2 times body weight, escalating to 2.5 times during more strenuous activities like running and jumping, the distribution and magnitude of the plantar pressure significantly influence foot health. [1]. The human foot comprises mainly bones and soft tissues, and due to inherent skeletal morphology differences and the influence of external forces during movement, the foot's bony structure, ligaments, and plantar soft tissues (PST) undergo elastic deformation, collectively determining the distribution of plantar pressure. Different regions accumulate varying stress over time, resulting in distinct deformations, reflected specifically in changes in foot arch morphology (FAM) and plantar soft tissue thickness (PSTT) [2]. Consequently, alterations in FAM and PSTT further impact the force direction and balance, thereby affecting foot health [3]. Therefore, the characteristics of FAM and PSTT are two vital factors extensively attended to by clinicians.
The FAM is defined by multiple indices and serves as a key indicator in clinical evaluation of various foot pathologies. For instance, the diagnosis of flatfoot and high-arch feet primarily relies on foot arch height [4], which is determined by the curvature of the longitudinal arch and transverse arch, and influences foot stability, shock absorption, and propulsion efficiency [5]. Radiographic imaging is usually used to evaluate the foot morphology under weight-bearing conditions when evaluating foot deformity clinically [6]. Particularly, weight-bearing lateral X-rays of the foot are considered the gold standard for diagnosing adult-acquired flatfoot deformity [7] and for assessing medial longitudinal arch characteristics [8].
The PST is subjected to the highest mechanical loads in the human body and has developed unique properties over time to accommodate these demands. These include wear resistance, pressure tolerance, and the limitation of interlayer displacement. Functionally, the PST supports weight, absorbs and cushions impacts from the ground, and maintains body balance and stability [9, 10]. It consists of complex structures such as skin, adipocytes, fascial layers, and muscles [11]. Variations in PSTT can influence the distribution of plantar pressure and the overall biomechanical behavior of the foot. Clinical observations and statistical data indicate that PST degeneration is linked to many common foot and ankle disorders, particularly in elder adults. Such degeneration may lead to pain and contribute to conditions like metatarsalgia, plantar fasciitis, hallux valgus, and complications in diabetic foot conditions [12–18].
While FAM and PSTT are common indicators for evaluating foot health clinically, the evaluation methods exhibit subjectivity and lack standardized criteria. In clinical practice, doctors typically manually annotate and measure radiological imaging results to assess FAM, focusing on parameters such as the calcaneal pitch angle (CPA) and the talo-first metatarsal angle (TMA or Meary's angle) to diagnose conditions like flatfoot and high arches [19]. However, these angles can be measured using approximately four to six different methods [20], and there is no standardized approach for defining and measuring arch height. Additionally, these manual measurements are time-consuming for practitioners.
Worse still, detecting variations in PSTT poses greater challenges. PSTT exhibits strong individualized characteristics, influenced by factors such as age, gender, diseases, and lifestyle [21][22], making standardized evaluation difficult. With the advancement of ultrasound technology, ultrasound has become the primary method for detecting PSTT [23][24][25]. Additionally, with the development of computer technology and advanced medical imaging techniques, studies employing visual image processing techniques, including deep learning and artificial intelligence, for detecting PSTT in foot medical imaging have been widely reported [26][27][28]. Although some literatures have conducted meta-analyses on deep learning image segmentation for PST, there is still a lack of analysis regarding the correlation between PSTT and factors such as gender, age, and footwear habits, especially based on large data samples for comprehensive statistical parameter information on PSTT. Furthermore, due to the difficulty in collecting medical data, previous studies have not analyzed the correlation between foot skeletal structure and PST based on large data volumes. Therefore, previous investigations have provided inconsistent and inconclusive results regarding the relationship between FAM and PSTT, without sufficient exploration into the impact of demographic factors such as gender and age on this relationship. However, it’s a meaningful task to explore the associations between FAM, PSTT, and different population categories, as well as their intrinsic correlations. This work is crucial for accurate and convenient foot health assessment and diagnosis, understanding foot biomechanical structures, disease prevention and rehabilitation, footwear and orthotic design, and foot modeling, etc.
To address the aforementioned challenges, our study amassed a substantial dataset of weight-bearing lateral foot X-ray images, a type for which there is currently no publicly available dataset. A set of 1497 images is retrospectively collected from the foot and ankle database of Huashan Hospital (Shanghai, China) spanning the last decade, with the personal info anonymized and ethic review approved. Utilizing deep learning image segmentation techniques, we preprocessed these images by adjusting grayscale, removing noise, and normalizing the images, enhancing the model's robustness, stability, and accuracy [29]. We then trained a deep neural network to perform precise segmentation of the first metatarsal (FM), talus (TA), calcaneus (CA), navicular (NAVI) bones, as well as the overall foot boundary. This approach enabled automated, standardized, and batch processing for precise computations of FAM and PSTT, thereby yielding significant time and cost efficiencies.
Furthermore, our research interest lies in analyzing the homogeneity and heterogeneity of large data samples, using data-driven approaches to discover patterns of similarity and dissimilarity among population groups. We further investigated the correlation between FAM and PSTT among different population groups. The section 2 describes the methodology, covering the data source, composition, and preprocessing of the dataset, as well as the development of deep learning image segmentation models and the evaluation metrics for FAM and PSTT. Section 3 presents the results, detailing the performance of the image segmentation models, the data results for FAM and PSTT, and the analysis of the correlation between FAM and PSTT across different demographic groups. Section 4 discusses the methodologies, results, and underlying hypotheses, followed by a summary and future perspectives of the study. The overall workflow of the study is illustrated in Fig. 1.