After the installation of the hardware and the software, we applied this portable device in a pilot clinical study, which was approved by the Medical Ethical Committee of Shengjing Hospital of China Medical University (approval No. 2020PS525K); however, no application has been filed for commercial use to the regulatory agencies. Our pilot study (ChiCTR.org.cn Identifier ChiCTR2000033830) aimed to investigate the usability and advantages of Auscul Pi in comparison with the 3M Littmann 3200 electronic stethoscope. To assess the auscultation performance of heart sounds and respiratory sounds with Auscul Pi, we included eight patients with structural heart disease or heart failure and two healthy volunteers, who were examined face-to-face with the device in the clinic or inpatient department. None of the participants had been diagnosed with SARS-CoV-2 infection. Patients and volunteers gave written informed consent in this study.

Inclusion criteria were (1) patients with New York Heart Association class IV heart failure, from whom rale sounds, including moist crackles and wheezes, could be heard in lung auscultation; (2) patients with any type of structural heart disease, such as congenital heart disease or valvular heart disease, from which murmurs could be heard. Patients were excluded if they had weak heart sounds caused by pericardial effusion, pleural effusion, or pneumothorax. Two healthy volunteers were also included in the study.

We divided the participants into three groups according to the inclusion criteria: (1) the respiratory sound group (rale group) contained patients with heart failure; (2) the heart sound group (murmur group) contained patients with structural heart disease; and (3) the healthy group (normal respiratory sound and heart sound group) contained healthy volunteers. The auscultation procedure was different for each group. To collect respiratory sound from the patients with heart failure in the rale group, we auscultated their left and right lungs in the regions of the 7th to 9th intercostal space (7ICS to 9ICS) along the midaxillary line. We first performed auscultation with Auscul Pi by pressing the “Auscultate” button on the touch screen to initiate the 30-second recording and broadcast. We checked the respiratory sounds from the microspeaker without ear contact. During that time, we assessed whether we could clearly hear moist crackles or wheezes at the bedside. Then, we repeated the procedure using the Littmann stethoscope.

Before auscultation of patients with structural heart diseases in the murmur group, we checked the echocardiogram to locate the main origin of the murmur; then, we focused on the corresponding site for auscultation. For example, the echocardiogram of one patient with valvular heart disease showed mild mitral regurgitation. Therefore, we auscultated at the apex site, located around the 5ICS in the midclavicular line, to hear the loudest murmurs from the mitral valve. In one patient with congenital heart disease, the echocardiogram showed a ventricular septal defect. We auscultated in the 3rd intercostal space (3ICS) and 4th intercostal space (4ICS) to the left border of the sternum to hear the loudest murmurs. We listened carefully to the output from the microspeaker of the Auscul Pi to assess the presence and clearness of the murmurs. Next, we checked healthy volunteers with both stethoscopes to evaluate normal heart sounds and respiratory sounds.

After auscultation, we transferred the data from the two stethoscopes onto a personal computer via Wi-Fi (Auscul Pi) or Bluetooth (Littmann 3200). First, we listened to the sound files (.wav) from both stethoscopes and compared the respiratory and heart sounds and their quality. Second, we compared the PCGs generated from each stethoscope with each other and with the electrocardiograms (ECGs) showing the cardiac cycles. To quantify the consistency of the two PCGs, we evaluated the relationship of the waveforms between the Auscul Pi and the Littmann stethoscope by assessing whether they had similar simultaneous ups and downs in the waveform and whether they showed similar S1, S2, and murmur timings.

For the analysis of respiratory sound auscultation, we used the audio data collected from the patients with heart failure, and we listened to the audio file from our Auscul Pi stethoscope to evaluate the consistency with the results obtained from the 3M Littmann stethoscope. For the heart sound auscultation analysis, in addition to listening to the audio, three physicians (WZ, XZ, and SG) compared the PCGs plotted from the digital array files from each stethoscope for the morphologies of the waveforms. Furthermore, a Pearson correlation analysis was performed to assess the consistency of the results obtained with the two stethoscopes. We first processed the PCG data by extracting the wave amplitude values at every time point as a data series. Then, the correlation between the two data series was implemented to evaluate the peak and trough synchronizations of S1, S2, and murmurs using Python code, and we also provided the source code in our GitHub repository [39].

Auscul Pi is modular to enable construction of the entire device in a short time. We required 4 weeks to design, purchase, and assemble the hardware and code, debug, and optimize the software of Auscul Pi after we had the initial idea. Due to the size (10 cm × 6 cm × 5 cm) and light weight, the Auscul Pi can be carried with a single hand, while the other hand holds the chest piece of the stethoscope for auscultation. The standby time of the batteries was 2.5 hours during the auscultation examination, and the batteries could be fully recharged in 2 hours via the mini-USB port on the uninterruptible power supply extension board. We found that we could operate it conveniently and record the information precisely in our clinical practice when considering the aspects of ergonomics and information technology.

To make data-based decisions about the prototype design, we evaluated the performance of the Auscul Pi based on the seven dimensions mentioned in the Methods. We used a scoring system with 5 levels of satisfaction, each of which was each scored from 1-5, with 5 being the strongest score (most satisfactory) and 1 being the weakest score (least satisfactory). The evaluators were WZ, XZ, and SG, who gave the scores. All scores were weighted by importance to obtain the total scores. The total score of Auscul Pi was 104, which was higher than that of the conventional stethoscope (87) and the 3M Littmann stethoscope (82) (Table 2).

This pilot study included eight patients and two healthy volunteers (Table 3). No patient was excluded from the study. First, we auscultated the two healthy volunteers to acquire normal heart sounds (Multimedia Appendix 4, Multimedia Appendix 5) and normal respiratory sounds (Multimedia Appendix 6, Multimedia Appendix 7). The audio of the respiratory and heart sounds obtained with Auscul Pi was clear and recognizable in both real-time play and the recorded archives. The digital NumPy array files of the corresponding sounds were used to plot the PCGs (Figure 7). The audio and PCGs generated by Auscul Pi were consistent with those obtained from the 3M Littmann stethoscope by the evaluations of the three physicians mentioned in the Methods section. To quantify the consistency of the two PCGs, we evaluated the relationship of the waveforms between the Auscul Pi and 3M Littmann stethoscopes by assessing whether they had similar peaks and valleys in the waveforms simultaneously, especially whether they had good synchronization of the first heart sound, second heart sound, and murmur timings. We firstly processed the PCG data by extracting the wave amplitude values of every time point. Then, we performed the Pearson correlation of the 2 data series, also with Python [39] (Multimedia Appendix 8). For the two healthy volunteers, the data for volunteer 9 had a correlation coefficient of 0.5570 (P<.001), and the correlation coefficient for volunteer 10 was 0.3245 (P<.001).

In the respiratory sound group (patients 1-4), all patients initially complained of dyspnea when presented at the hospital, and we heard rales in all patients. All these patients received anti-heart failure therapies. They all recovered and were discharged several days later. Patient 1 presented with atrial fibrillation and heart failure. At the beginning of the treatment period, we performed auscultation on this patient with both stethoscopes. We could hear clear wheezes in both the inspiratory and expiratory phases. During the examination, we checked the respiratory sounds simultaneously played by the microspeaker, from which the wheezes were clear and obvious. Then, we replayed the wheezing sounds in the computer when away from the patient, and the wheeze sound quality and recognizability were better than those obtained when we broadcast the sound during recording (Multimedia Appendix 9). The quality of the respiratory sounds was good for the other three patients in the group (patients 2-4) during playback of the recording.

For the patients in the heart sound group (patients 5-8), we examined and easily detected the murmurs at the corresponding auscultation sites of the culprit valves or defects. For example, patient 8, who was suffering from congenital heart disease, had two intraventricular septal defects. When we auscultated him at 3ICS and 4ICS to the left border of the sternum, a loud holosystolic murmur was clearly detected with Auscul Pi (Multimedia Appendix 10). The acoustic characteristics and timings of the murmurs were quite similar to the murmurs heard with the Littmann stethoscope (Multimedia Appendix 11). This patient underwent surgical ventricular septal repair. Postsurgery auscultation with the Auscul Pi (Multimedia Appendix 12) and Littmann stethoscope (Multimedia Appendix 13) stethoscopes showed that the murmurs had disappeared.

The alignments of the PCGs with the ECGs showed good visual consistency between Auscul Pi and the 3M Littmann stethoscope (Figure 8). We also performed the same correlation analysis to evaluate the consistency of the data series extracted from the PCG NumPy data. The correlation coefficient of the Auscul Pi and 3M Littmann results before surgery was 0.3436 (P<.001), and the coefficient after surgery was 0.5138 (P<0.001). The correlation coefficients of the other 3 patients ranged from 0.3449-0.4797 (P<.001) (Table 4).

In this work, we present the development of Auscul Pi, an innovative electronic stethoscope that we evaluated in a pilot study in patients with cardiovascular diseases. Our results show that Auscul Pi can be applied for the examination of cardiovascular diseases, as it clearly plays and records heart and respiratory sounds. The consistency between the results obtained by Auscul Pi and a typical electronic stethoscope was dependable based on qualitative analysis of the audio files and statistical analysis of the PCGs.

Our team found that Auscul Pi had several advantages over traditional stethoscopes in clinical practice. First, the contactless design of the stethoscope allowed no contact with the ears of the medical staff. The device can be used by physicians and nurses wearing a protective suit, eye protector, and face shield when auscultation is essential in clinical work, such as treating patients with COVID-19. The stethoscope can reproduce respiratory and heart sounds with the microspeaker, enabling nearby medical staff to hear the broadcast as well as if they were touching their ear to a stethoscope earpiece.

Second, the components are inexpensive, and component assembly and software installation are relatively easy; therefore, the do-it-yourself protocol can be followed by medical staff in a short time. The cost of the entire device is approximately US $94, of which $35 correspond to the Raspberry Pi; this is much lower than the price of a 3M Littman stethoscope. Therefore, the device is accessible to most medical facilities, hospitals, and emergency rooms worldwide. Additionally, Raspberry Pi is a popular project worldwide. It can be easily ordered on the web, and due to its small size, fast shipment is possible.

Third, in addition to its low cost, Raspberry Pi has versatility for application in many medical projects, ranging from assistance with medical imaging [40] to cervical cancer prevention [41], building computational microscopy [42], and ventilator buildup during the COVID-19 pandemic [42,43]. Moreover, Raspberry Pi uses Linux as a routine operating system, which is open-source, generic, and freely downloadable. Additionally, many software packages developed by third-party developers use Python, which is one of the most popular programming languages and works well with Raspberry Pi.

Fourth, the recorded respiratory and heart sound data can be stored in the Raspberry Pi, then transferred to a personal computer for further analysis if the hospital, emergency department, ICU, or ward has a Wi-Fi signal.

Finally, Auscul Pi can quantify and visualize auscultation: the system simultaneously records and broadcasts the signal, and the resulting computer files can be transferred using Wi-Fi for offline analysis. PCGs can also be plotted based on the NumPy array files for ease of visualization and analysis. These data can also be used in the future for research related to diagnosis and prognosis, such as use of machine learning algorithms [24,44]. The device may also have educational value as a teaching aid for medical students.

While treating patients with COVID-19 in Wuhan, we developed the Auscul Pi to solve the problems of auscultation. In a long-term perspective, this innovation may not be limited to the examination of patients with COVID-19, and it may be applied to other infectious diseases to reduce the risk of infection of medical workers. Although we do not envisage that Auscul Pi will become a commercial medical product in a large market, we believe it may inspire biomedical engineers, bioinformatics researchers, clinicians, and computer scientists to create low-cost engineering technologies to benefit patients who have been severely affected by COVID-19. Low-cost portable medical devices based on Raspberry Pi and the Python programming language may even become useful as tools for self-monitoring and assessment by patients with COVID-19 under quarantine [45], especially in low-resource areas.

The auscultation sounds recorded and broadcast by Auscul Pi inevitably contain some noises due to background and electricity, such as tiny click and pop sounds, that nevertheless do not cover up the main auscultation sounds. Future work should focus on filtering out much or all of this background noise. Our clinical research was a small pilot study to explore the feasibility of using the device in patients; however, we have planned a randomized clinical trial involving more medical professionals as device users and more patients with auscultatory abnormalities. We will also use questionnaires and unstructured interviews to ask the professionals about the usability and reliability of the device.

A low-cost electronic stethoscope device, Auscul Pi, enables auscultation without ear contact. The device enables real-time broadcast of auscultation sounds and simultaneous digital data storage for offline analysis. Auscul Pi may enable accurate auscultation of patients with COVID-19 by medical workers wearing protective suits, thereby helping to minimize risk of infection.