A High‐Accuracy and Easy‐Use Differential Diagnosis Method of Bipolar Disorder Depressive Phase and Major Depressive Disorder Using a Wearable Near‐Infrared Spectroscopy

ABSTRACT

Objectives

Although conventional multichannel near‐infrared spectroscopy (cNIRS) has been reported to be useful in differentiating bipolar disorder depressive phase (BDD) from major depressive disorder (MDD), its utilization in real clinical practice remains limited. The objective of this study was to evaluate the performance of wearable NIRS (wNIRS), which is cheaper to install and easier to use than cNIRS, in differentiating between BDD and MDD, and to propose an optimal index for wNIRS.

Methods

The subjects were 16 BDD patients, 16 MDD patients, and 16 healthy controls. Changes in the total hemoglobin signal (Δ[total‐Hb]) in the bilateral frontal cortex during a 60 s verbal fluency task (VFT) were measured using the HOT‐2000, a wNIRS capable of reducing noise from skin blood flow. In addition to evaluating the discriminative performance of the integral and centroid values proposed in previous studies, the ratio of the mean values of the Δ[total‐Hb] during the first 20 s and subsequent 40 s (20–40 ratio) during the VFT was also employed and assessed.

Results

Only the 20–40 ratio showed significant differences between BDD and MDD, whereas the integral and centroid values did not. Using a cut‐off value of 2.00 for the 20–40 ratio, the sensitivity and specificity of the BDD diagnosis were 100% and 93.4%, respectively.

Conclusions

The wNIRS measurements and the 20–40 ratio of the Δ[total‐Hb] in the forehead area enabled more accurate differentiation between MDD and BDD than the conventional analysis.

1. Introduction

The medication of bipolar disorder depressive phase (BDD) involves the use of mood stabilizers and antipsychotics [1], which is different from the use of antidepressants in major depressive disorder (MDD). The use of antidepressants for bipolar disorder can make the patient's mood more unstable [2, 3, 4], including induction of mania and rapid cycling, and may contribute to increased suicide rates. Antidepressants should generally be avoided in bipolar disorder. However, approximately 20% of patients with an initial diagnosis of “MDD” present with a manic episode several years after the initial depressive episode and the diagnosis is changed to “bipolar disorder,” [5, 6] suggesting that a large number of BDD patients may receive inappropriate treatment. Furthermore, bipolar disorder is known to worsen the longer the patient is not properly treated as “bipolar,” [7, 8] so a correct diagnosis in the early stages of the disease is very important.

Clinical psychiatric diagnoses are based on the Diagnostic and Statistical Manual of Mental Disorders‐5 (DSM‐5) [9] and the International Statistical Classification of Diseases and Related Health Problems‐10 (ICD‐10) [10]. However, it is difficult to differentiate bipolar disorder from MDD in the early stages of the disease because bipolar disorder often begins with depression rather than mania [1, 11]. Although frequent symptoms are shown for each of BDD and MDD [12, 13], they are not enough to lead to a clear differentiation in real clinical practice. There are also reports of head magnetic resonance imaging (MRI) and positron emission tomography (PET) studies of the findings [14, 15], but they are not specific to each disease. In the first place, it is difficult for general psychiatric hospitals to install such large‐scale diagnostic imaging equipment.

In previous studies, conventional multichannel near‐infrared spectroscopy (cNIRS) has been used as an aid in differential diagnosis of the depressive state, and shown characteristic waveforms of BDD and MDD using a verbal fluency task (VFT) [16]. In NIRS research, the VFT is often used as a cognitive task which examines frontal cortical and cognitive function in humans [16, 17, 18, 19, 20, 21], because it is easy for research subjects to understand, is not invasive, can be performed in a wide range of age groups, has good task responsiveness, and has shown waveform differences in diseases. The VFT comprises two types of tests, namely, the letter fluency test (LFT) and word fluency test (WFT). The LFT uses three letters (e.g., /a/, /to/, /na/) as the first syllable to produce as many words orally as possible within 60 s (sec). Three letters are used to change the initial letter every 20 s to make the response as uninterrupted as possible. The WFT requires subjects to generate verbally as many words as possible within 60 s according to three categories (e.g., /animals/, /plants/, /vehicles/). The LFT mainly involves the frontal cortex and the WFT mainly involves the temporal cortex [22]. When changes in the oxygenated hemoglobin signal (Δ[oxy‐Hb]) in the frontal cortex were examined by cNIRS using the VFT, the peak in the early phase of the VFT was smaller in MDD and there was little increase thereafter [16]. For MDD, cNIRS has also confirmed that the frontal function continues to decline even when the disease is in remission [23]. Although BDD shows a greater VFT‐induced Δ[oxy‐Hb] increase, it is significantly less active than normal subjects [24], with a delayed initial increase and a peak in the latter half of the task [16, 17]. On the other hand, some studies have shown other waveform characteristics, and disease‐specific waveforms have not been established yet [25]. However, previous studies have demonstrated the clinical utility of cNIRS as an “adjunct diagnosis” in the differentiation of psychiatric disorders resulting in depressive state [18].

Compared with MRI and PET, cNIRS is small, mobile, and does not require strong instrumental restraints (e.g., maintenance of the supine position for several minutes). However, it is still not small enough to be portable and has required patients to make appointments at specific medical facilities for testing. In other words, many patients visit hospitals with a depressive state, but real‐time cNIRS testing was difficult when psychiatrists were struggling to make a diagnosis. This was a disadvantage for both the patients and the psychiatrists.

On the other hand, wearable NIRS (wNIRS) is very compact (e.g., the portable brain activity measurement device used in this study, HOT‐2000 (NeU Corporation), has dimensions W225 mm × D125 mm × H51 mm and weighs 129 g) and is inexpensive. It is easy to attach to the patient and can be handled at any medical institution. It can be placed anywhere and can be easily moved from one room to another. Furthermore, measurements can be taken in a natural posture.

In addition, superimposition of physiological artifacts such as scalp blood flow has been considered problematic. Therefore, the accuracy of the modified Beer–Lambert law analysis by cNIRS using conventional continuous light (1 light‐transmitting −1 light‐receiving type) has been questioned [26, 27]. One method of reducing scalp blood flow is the multiple probe placement method [28, 29]. In NIRS measurements, a light‐receiving probe is generally placed at a distance of 3.0cm from the light source. This method adds a light‐receiving probe, which is placed at a distance of 1.0cm from the light source. This enables the separation of deep tissue–derived components from shallow tissue–derived components [30]. The HOT‐2000 uses the real‐time scalp signal separating algorithm of Kiguchi and Funane [29], and evaluates changes in the Δ[total‐Hb], and has also been tested for clinical applications [31].

Should the HOT‐2000 demonstrate a similar or enhanced capacity for discriminative analysis compared with cNIRS, it would offer a significant advantage in terms of its user‐friendly design. In this study, the discriminative performance of the HOT‐2000 for BDD and MDD was evaluated.

2. Materials and Methods

2.1. Participants

This study was designed at the Graduate School of Medicine, Kyoto University, and only measurements were made with the cooperation of Soujinkai Chikugoyoshii Cocorohospital (hereafter referred to as Cooperating Hospital). Approval was obtained from the Ethics Committee of the Graduate School of Medicine and Faculty of Medicine of Kyoto University and from the Ethics Committee of the Cooperating Hospital. Furthermore, in accordance with the Declaration of Helsinki, all participants gave their written informed consent after receiving a full explanation of the research.

Sixteen inpatients or outpatients treated for MDD (mean age ± standard deviation [SD] = 41.3 ± 14.0 years, 20–69 years), 16 inpatients or outpatients treated for BDD (48.0 ± 13.4 years, 23–69 years), and 16 healthy controls (HC) (42.3 ± 14.0 years, 21–64 years) were enrolled (see Table 1). All participants were native Japanese speakers and could easily perform the VFT.

TABLE 1.

Characteristics.

	HC	MDD	BDD
Subject number	16	16	16
Female/male	9/7	9/7	10/6
	Mean ± SD	Mean ± SD	Mean ± SD
Age (years)	42.3 ± 14.0	41.3 ± 14.0	48.0 ± 13.4
VFT ko	5.1 ± 1.7	5.1 ± 2.2	4.6 ± 1.8
ka	5.7 ± 1.8	5.3 ± 1.6	5.2 ± 1.1
ku	4.8 ± 2.1	5.6 ± 2.3	5.1 ± 1.3
VFT total	15.6 ± 5.7	16.0 ± 5.3	14.9 ± 3.6
HAMD a score	—	10.9 ± 3.8	9.6 ± 1.9
Degree Normal/mild/ Moderate/severe	—	1/11/2/0	1/12/1/0

Abbreviations: BDD, bipolar disorder depressive phase; HAMD, Hamilton Rating Scale for Depression; HC, healthy control; MDD, major depressive disorder; VFT, verbal fluency task.

^a HC has no data for HAMD. HAMD evaluation was performed in 14 of 16 for both MDD and BDD subjects. There was no significant difference. Scores of 0–7 were classified as normal, 8–13 mild, 14–18 moderate, and 19– severe.

The patients were diagnosed by psychiatry specialists according to the DSM‐5. HCs were self‐reported staff members of the cooperating hospital who had never had a depressive mood or loss of interest lasting more than 2 weeks and had never received psychiatric treatment.

The exclusion criteria for participants were comorbidity of any of the following: psychotic disorders (e.g., schizophrenia), organic diseases (e.g., neurological illness, traumatic brain injury, dementia), alcohol/substance addiction, or endocrine diseases (e.g., hypothyroidism).

On the day of the NIRS measurement, depressive symptoms of the patient subjects were assessed using the 17‐item Hamilton Rating Scale for Depression (HAMD) [32].

Some patient subjects were taking their medications; others were not. There have been few studies that have found a significant association between NIRS data and psychotropic medication dosage. A few reports on the differentiation of MDD, BDD, and schizophrenia using NIRS indicate that the results examined in the unmedicated subjects are similar to those in the medicated subjects [16, 33]. Therefore, we decided not to consider the patient subjects with regard to their medications.

2.2. Verbal Fluency Tasks

Of the two types of the VFT, the LFT was used as the activation task because the HOT‐2000 measures the [total‐Hb] in the frontal cortex. Each trial consisted of a 60 s pre‐VFT control period (task1), a 60 s VFT period (task2), and a 60 s post‐VFT control period (task3), and was a single trial. During the task1 and task3, the subjects were requested to repeatedly verbalize five Japanese vowels (/a/,/i/,/u/,/e/, and/o/). This was intended to correct the data during the task2 for activation due to vocalization. During the task2, the subjects were requested to verbalize as many words as possible that begin with one Japanese initial they were instructed to speak. The initials were/ko/,/ka/, and/ku/. These are the most frequently used sounds at the beginning of Japanese words, with a large number of Japanese words listed in the Sanseido Shinmeikai Kokugo Dictionary (Japanese Dictionary).

2.3. NIRS Measurement

A headband‐style detector of the HOT‐2000, a 1 light‐transmitting and 2 light‐receiving 2‐channel probes, was placed horizontally along the T3‐Fpz‐T4 line of the international EEG 10–20 system. (Fig. 1). The light‐receiving probes were positioned at the Fp1 and Fp2 sites of the subjects (Fp1 is shown hereafter as left side and Fp2 as right side), and the light‐transmitting position was 3.0 cm outside of the light‐receiving probes. The reference light‐receiving probe was positioned at a distance of 1.0 cm. With this arrangement, the [total‐Hb] in the frontal cortex was measured. The instrument measures the [total‐Hb] (mMmm) at a sampling interval of 0.1 s using a wavelength of 810 nm near the isosbestic point of the oxy‐Hb and the deoxygenated Hb (deoxy‐Hb). The measurement data were transmitted and saved in real time via Bluetooth to an attached tablet (LAVIE PC‐TE510JAW, NEC).

FIGURE 1.

Schematic of wNIRS probe configuration and anatomical placement. The device is oriented symmetrically along the T3–Fpz–T4 line (International 10–20 system). Blue rectangles: light‐receiving detectors at Fp1 and Fp2; red rectangles: light‐emitting sources placed 3.0 cm laterally to detectors; green rectangles: reference detectors offset by 1.0 cm from sources to monitor scalp blood flow.

2.4. Signal Processing

To reduce short‐term motion artifacts and high‐frequency noise such as pulse waves caused by heartbeats in the collected data, a 3 s moving average was applied to the 0.1 s sampled data of the [total‐Hb].

The pre‐VFT and the post‐VFT baselines were the average of the last 10 s of the task1 and the last 5 s of the task3, respectively, according to the previous study reported by Takizawa et al. [34] Linear fitting was performed on the data from these two baselines to calculate the Δ[total‐Hb]. The analysis interval was also based on previous studies as {(last 10 s of the task1) + (60 s of the task2) + (55 s of the task3)} [16, 34].

We also obtained the integral and centroid values in the analysis interval for subjects (Figs. 2A and 2B), according to the method of Takizawa et al. [16] The integral value represents the magnitude of the hemodynamic response during the VFT and the centroid value represents the timing of the hemodynamic response. The integral value (arbitrary units: a.u.) is the area under the curve (AUC) of the Δ[total‐Hb] during the task2. The centroid value (sec) is the time indicated by the vertical line from the centroid of the Δ[total‐Hb] region (calculated as a positive change) throughout the analysis interval. In addition, we obtained the ratio of the mean value of the Δ[total‐Hb] in the first 20 s to the subsequent 40 s (20–40 ratio) during the task2 for subjects. We created a receiver‐operating characteristic curve (ROC) using the left and right 20–40 ratios and examined possible cut‐off values (COs).

FIGURE 2.

Schematic description of how centroid and integral values are calculated. The typical time‐course pattern of the Δ[total‐Hb] at the analysis interval (pre‐VFT control period 10 s + VFT period 60 s + post‐VFT period 55 s) is illustrated in both (A) and (B). The centroid value is defined as the time T at which areas A and A' have the same area throughout the analysis interval, as illustrated in (A). The integral value represents the sum of the positive and negative values of the intervals during which the VFT is being performed, as shown in (B). Δ[total‐Hb]: changes in the total hemoglobin signal; VFT: verbal fluency task.

2.5. Statistical Analyses

The Kolmogorov–Smirnov test was used to ensure that the age and VFT performance scores were normally distributed across HC, MDD, and BDD subjects. The Bartlett test was used to ensure equal variances in the age and VFT performance scores among HC, MDD, and BDD subjects. We performed multiple tests of means (one‐way analysis of variance with Bonferroni correction, significance level: α = 0.05) for the age and the numbers of VFT performances, respectively.

Similarly, the Kolmogorov–Smirnov test was used to ensure that the integral values, the centroid values, and the 20–40 ratios were normally distributed across HC, MDD, and BDD subjects. The Bartlett test was used to ensure that the integral values, the centroid values, and the 20–40 ratios had unequal variances among HC, MDD, and BDD subjects. We performed multiple tests of means (Welch's t‐test with Bonferroni correction, significance level: α = 0.05) for the integral values, the centroid values, and the 20–40 ratios, respectively.

We ensured that HAMD scales of MDD and BDD subjects showed normal distributions and were of unequal variance by F‐test, so Welch's t‐test (significance level: α = 0.05) was performed to compare the HAMD scales between MDD and BDD subjects.

Statistical analysis and ROC curves were conducted with EZR version 1.63 software [35].

3. Results

3.1. Subjects Characteristics

Sex, age, the numbers of VFT performances, and the HAMD evaluation among HC, MDD, and BDD subjects are shown in Table 1. There were no significant differences in sex, age, and the numbers of VFT performances among HC, MDD, and BDD subjects. The HAMD evaluation was performed in 14 of 16 for both MDD and BDD subjects. There was no significant difference.

3.2. Characteristics of Waveform of NIRS Signals

By visual inspection of the Δ[total‐Hb] curve shapes, in 14 of 16 HC subjects, the Δ[total‐Hb] rose steeply at the beginning of the task2 in both the left and right sides, remained at an activated level with further increase in the latter half during the task2, and fell quickly after the task2. In 11 of the 16 MDD subjects, the Δ[total‐Hb] rose steeply at the beginning of the task2 in both sides, but reached a plateau immediately, and fell quickly after the task2. In 11 of 16 BDD subjects, the Δ[total‐Hb] rose more slowly than the Δ[total‐Hb] of HC and MDD subjects during the beginning of the task2 in both sides, with a delayed peak of activation. The peaks were highest in HC and lowest in MDD. The average waveforms of the Δ[total‐Hb] change for 16 HC, 16 MDD, and 16 BDD subjects, respectively, were shown in Figs. 3A and 3B. Waveform patterns were similar to those previously reported using the Δ[oxy‐Hb] in the frontal region by cNIRS [24], though the waveform of BDD subjects in this study tended to be more slowly elevated during the task2.

FIGURE 3.

Averaged Δ[total‐Hb] waveformsa cross groups. Waveforms (0.1 s resolution) are shown for healthy control (HC), major depressive disorder (MDD), and bipolar disorder depressive phase (BDD) (n = 16 per group) in the (A) left and (B) right sides. Shaded regions denote ±SD. The vertical red line at 30 s marks the division of the VFT period (first 20 s vs. subsequent 40 s). BDD shows a bilaterally delayed hemodynamic response at VFT onset compared to HC and MDD. Δ[total‐Hb]: changes in total hemoglobin signal. VFT: verbal fluency task.

3.3. Conventional Index: Integral and Centroid Values

When expressed as mean ± SD, on the left side, the integral values (a.u.) were 15.0 ± 6.7 for HC subjects, 3.6 ± 2.8 for MDD subjects and 5.3 ± 5.7 for BDD subjects, and the centroid values (sec) were 49.4 ± 4.9 for HC subjects, 48.3 ± 19.6 for MDD subjects and 59.9 ± 17.2 for BDD subjects. On the right side, the integral values were 15.0 ± 8.4 for HC subjects, 3.9 ± 5.1 for MDD subjects, and 6.9 ± 5.0 for BDD subjects, and the centroid values were 47.8 ± 6.3 for HC subjects, 41.2 ± 16.8 for MDD subjects, and 54.5 ± 11.9 for BDD subjects.

On both sides, significant differences were found in the integral values between HC and MDD subjects (both sides: p < 0.0001), and significant differences were also found between HC and BDD subjects (left side: p < 0.001, right side: p < 0.01), but there was no significant difference between MDD and BDD subjects. No significant difference was found among HC, MDD, and BDD subjects for the centroid values on both sides.

3.4. A New Index: 20–40 Ratio

When expressed as mean ± SD, on the left side, the 20–40 ratios were 1.2 ± 1.1 for HC subjects, 1.2 ± 0.5 for MDD subjects, and 3.7 ± 1.5 for BDD subjects. On the right side, the 20–40 ratios were 1.5 ± 0.6 for HC subjects, 1.2 ± 0.8 for MDD subjects, and 3.5 ± 2.0 for BDD subjects. The box‐and‐whisker plots of the 20–40 ratios for HC, MDD, and BDD subjects are shown in Figure 4A and 4B.

FIGURE 4.

Comparison of the 20–40 ratio during the verbal fluency task. Ratios for the first 20 s to the subsequent 40 s are shown for healthy control (HC), major depressive disorder (MDD), and bipolar disorder depressive phase (BDD) groups. Significant differences were observed between BDD and both HC and MDD on the (A) left and (B) right sides. No significant differences were found between HC and MDD on either side. Δ[total‐Hb]: changes in total hemoglobin signal.

On both sides there were significant differences between HC and BDD subjects (left side: p < 0.0001; right side: p < 0.01) and between MDD and BDD subjects (left side: p < 0.0001; right side: p < 0.01). Figure 5A and 5B plotted the integral values and the 20–40 ratios in HC, MDD, and BDD subjects. It can be posited that these two indicators can be employed in conjunction to effectively differentiate among HC, MDD, and BDD subjects. In contrast, the plotting of the centroid value and the 20–40 ratio did not demonstrate such a clear differentiation between the three groups.

FIGURE 5.

Scatter plots of the 20–40 ratio and integral values. (A) Left‐side and (B) right‐side plots of the integral value and the 20–40 ratio. The combination of these two indicators effectively differentiates healthy control (HC), major depressive disorder (MDD), and bipolar disorder depressive phase (BDD) groups. In contrast, the combination of the centroid value and the 20–40 ratio (not shown) does not provide equivalent diagnostic separation. the ratio of the mean value of the 20–40 ratio: the ratio of the mean value of the Δ[total‐Hb] in the first 20 s to the subsequent 40 s.

ROC curve analysis was used to determine the optimal CO of the 20–40 ratios for the differential diagnosis of BDD and MDD subjects by identifying the extreme top left point of the ROC curve. The CO and the AUC for the left side (Figure 6A) were calculated to be 2.170 and 0.975 with 95% confidence interval (CI) of 0.932–1, whereas the CO and the AUC for the right side (Figure 6B) were 1.790 and 0.912 with 95% CI of 0.810–1.

FIGURE 6.

ROC curve analysis of the 20–40 ratio in bipolar disorder depressive phase (BDD) and major depressive disorder (MDD). (A) Left‐side performance: cut‐off, 2.170; AUC, 0.975 (95% CI, 0.932–1.000). (B) Right‐side performance: cut‐off, 1.790; AUC, 0.912 (95% CI, 0.810–1.000). The 20–40 ratio is the ratio of the Δ[total‐Hb] mean values in the first 20 s to the second 40 s. AUC: area under the curve; CI: confidence interval.

Using the above calculated CO values on each side, 15 out of 16 BDD subjects and 15 out of 16 MDD subjects were correctly classified on the left side, and 15 out of 16 BDD subjects and 13 out of 16 MDD subjects were correctly classified on the right side.

If we diagnosed BDD in the differentiation between MDD and BDD when either side exceeds 2.00, which is an integer between the left and right CO, the results show that all 16 BDD and 15 of 16 MDD subjects could be correctly determined, which meant the sensitivity and specificity of BDD diagnosis between two diseases were 100% and 93.4%, respectively.

4. Discussion

We investigated the ability to discriminate between MDD and BDD subjects using wNIRS, which measures the [total‐Hb] in the frontal cortex. The cNIRS, which measures the oxy‐Hb and the deoxy‐Hb signal, has been employed as an auxiliary diagnostic tool in the assessment of psychiatric disorders in depressive states. The centroid values derived from frontal cortex measurements have been identified as a significant variable for differential diagnosis [16]. Subsequent studies and reports on the waveform characteristics, the integral and centroid values of psychiatric disorders that result in depressive states have continued [25, 36]. In this study, the Δ[total‐Hb] curve shape in the frontal cortex measurements showed similar characteristics to previous cNIRS studies using the Δ[oxy‐Hb] [16, 34]. This is consistent with previous reports that the Δ[oxy‐Hb] and the Δ[total‐Hb] are approximately equivalent [37, 38].

However, in this study, the Δ[total‐Hb] in BDD subjects exhibited a tendency to increase at a slower rate during the VFT. Several causes are considered. First, a few reports have suggested that the Δ[total‐Hb] has higher spatial specificity than the Δ[oxy‐Hb] [39, 40], and it is possible that the waveform characteristics were more prominent in the measurement sites in this study. Second, the integration of the multiple probe placement method into the HOT‐2000 may have resulted in higher accuracy than cNIRS measurements without similar noise reduction features, due to the reduction of noise associated with skin blood flow. A third potential explanation is the differing severity of the subjects. The majority of BDD subjects in this study exhibited mild HAMD ratings, whereas previous cNIRS reports may have included more moderate and severe ratings [16]. It is possible that the waveform characteristics of BDD patients may vary according to severity, which may have contributed to the positive outcomes observed in this study. Nevertheless, as mentioned in the introduction of this report, bipolar disorder is known to worsen the longer the patient is not properly treated as “bipolar.” [7, 8] Early diagnosis and early treatment at the time of mild disease level are important, and it is significant that we were able to accurately differentiate MDD and BDD subjects at the mild disease level in this study.

The integral values of the Δ[total‐Hb] obtained in this study and the integral values of the Δ[oxy‐Hb] from several previous studies [16, 34] both demonstrated significant differences between HC and BDD, as well as between HC and MDD. In contrast, no significant differences were observed between MDD and BDD in either set of findings. A study by Husain et al. using cNIRS found that during the VFT, MDD showed little to no increase in Δ[oxy‐Hb], exhibiting a significant difference in the integral value compared with bipolar disorder, which did show an increase [41]. This finding is at odds with the results of our study and those of previous reports. As was also mentioned in the text, the difference may be due to the use of an alphabet‐based language versus a syllabary like Japanese or Chinese. Furthermore, the higher severity of MDD on the HAMD score in their study may have contributed to the findings.

On the other hand, comparison of the centroid values of the Δ[total‐Hb] among HC, MDD, and BDD subjects revealed no significant differences, despite previous studies indicating that this variable may be a useful diagnostic tool in BDD. Therefore, we considered it necessary to have an index that can accurately differentiate between BDD and MDD, which are not centroid values, in wNIRS measurements. Thus, we proposed the 20–40 ratio as a new differential index in this study. Because a more gradual increase of the Δ[total‐Hb] in the BDD subjects was noticeable, we found significant differences between BDD and MDD subjects, and BDD and HC subjects, using the 20–40 ratio as an indicator. We calculated 2.17 for the left and 1.79 for the right as optimal values for BDD and MDD subjects. In this study, we did not measure special locations such as the language area, and organic disease was also an exclusion criterion for participation, so we considered functional and anatomical left–right differences to be scant, and we proposed a differential diagnosis of BDD if the 20–40 ratio of either the left or right side exceeds 2.00 as a threshold value. This method has two possible advantages. One is that it is easy to use because it does not require two thresholds for clinical use. Another is that this method increases the robustness of the differential diagnosis method. This takes advantage of the fact that both left and right data behave in the same way, making it possible to diagnose even if one of the data is not accurate for some reason.

This study investigated the ability of wNIRS (HOT‐2000) to discriminate between BDD and MDD, and proposed a new index. We believe that wNIRS and the new index in clinical diagnosis have the potential to significantly improve the ability to differentiate BDD from MDD at an early stage and may have contributed an important step toward solving the current problem in psychiatry of the lack of diagnosis based on objective measurements. In the future, the clinical usefulness of this diagnostic method should be verified by using it in a larger number of patients. The influence of various patient backgrounds, for example, age, sex, severity of illness, and medication on the accuracy of differentiation, as well as the influence of changing NIRS equipment, should also be investigated. As mentioned earlier, it should be noted that although medication has been reported to have no significant effect on NIRS results, there are a few reports that mention a possible influence [42, 43].

In recent years, NIRS and MRI have been used to elucidate the pathophysiology of bipolar disorder. Lim et al. demonstrated significant differences in functional connectivity between HC and bipolar disorder by observing the prefrontal language network with 48 NIRS detectors placed on the forehead [44]. Similarly, a study by Romeo et al. using functional MRI showed differences in functional connectivity across a broader brain region, including the frontal lobe, between HC and bipolar disorder [45]. These findings support the notion that NIRS measurements confined solely to the prefrontal cortex can detect significant differences between the two groups. While it is difficult to discuss brain functional connectivity or pathophysiology based solely on our results, such reports suggest that employing a simple wNIRS device worn only on the forehead for disease differentiation is a valid approach.

Given the convenience of wNIRS, we postulated that wNIRS would be clinically useful even if its performance were equivalent to that of cNIRS. Therefore, the performance of wNIRS, HOT‐2000, was evaluated for BDD and MDD. As a result, it is notable that the wNIRS and the novel analysis method proposed in this study demonstrated high differential diagnostic accuracy. The new method may also be applied to data obtained via cNIRS, potentially yielding enhanced outcomes. It is anticipated that the accurate differential diagnosis enabled by NIRS will result in a significant change in the utilization of objective diagnostic methods based on measurements with equipment for clinical testing in the future clinical practice of psychiatric disorders.

5. Conclusions

We proposed a new differential diagnosis method for BDD and MDD using wNIRS, which is inexpensive to install, easy to move and attach to the patient, and expected to have high differential diagnostic performance.

Author Contributions

All contributors planned the experiment, Yuji Sasaki measured the subjects at the Cooperating Hospital, and all contributors analyzed the data and discussed the results. Yuji Sasaki wrote the manuscript, and Koichi Ishizu, Akitoshi Seiyama, and Naozo Sugimoto supervised and corrected the manuscript. All contributors reviewed and approved the final version of the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Sasaki Y., Ishizu K., Seiyama A., and Sugimoto N., “A High‐Accuracy and Easy‐Use Differential Diagnosis Method of Bipolar Disorder Depressive Phase and Major Depressive Disorder Using a Wearable Near‐Infrared Spectroscopy,” Bipolar Disorders 28, no. 2 (2026): e70089, 10.1111/bdi.70089.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.